KR101117928B1 - Plasma process diagnosis system, method and apparatus of detecting an end point in the same - Google Patents

Abstract

The present invention relates to a plasma process diagnostic system and an endpoint detection method and apparatus therein. The endpoint detection method may include setting a first segment value corresponding to a first state and a second segment value corresponding to a second state, and process OES data based on the first segment value and the second segment value. Converting to a state function probability distribution and detecting an endpoint using a maximum likelihood state sequence in the state function probability distribution. Here, the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability.

Description

Plasma Process Diagnosis System and Method and Apparatus for Detecting End Point in It {PLASMA PROCESS DIAGNOSIS SYSTEM, METHOD AND APPARATUS OF DETECTING AN END POINT IN THE SAME}

The present invention relates to a plasma process diagnostic system and an endpoint detection method and apparatus therein.

The plasma process diagnosis system is a system for diagnosing an abnormality and a process state of a plasma process that performs deposition and etching processes using plasma, and detects an endpoint in, for example, a wafer etching process.

In general, the plasma process diagnosis system uses a slope of data to detect the end point, and when the size of the semiconductor is large, the end point can be accurately detected.

At present, the size of semiconductors is becoming much smaller. However, when the method of detecting an end point using the slope of the data is applied to a miniaturized semiconductor, there is a problem that the end point is not accurately detected.

That is, the conventional diagnostic system could not accurately detect the presence or absence of an abnormality in the plasma process and the process state.

SUMMARY OF THE INVENTION An object of the present invention is to provide a plasma process diagnosis system capable of accurately detecting a plasma process abnormality, a process state, and the like in a large semiconductor as well as a small semiconductor, and a method and apparatus for detecting an end point in the same.

In order to achieve the above object, the endpoint detection method according to an aspect of the present invention comprises the steps of setting a first segment value corresponding to the first state and a second segment value corresponding to the second state; Converting process OES data into a state function probability distribution based on the first segment value and the second segment value; And detecting an endpoint using the maximum likelihood state sequence tracking method in the state function probability distribution. Here, the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability.

An endpoint detection apparatus according to another aspect of the present invention includes a regression model unit for constructing a regression model to obtain a first segment value corresponding to a first state and a second segment value corresponding to a second state; A state function probability unit for converting process OES data into a state function probability distribution based on the first segment value and the second segment value; And an endpoint determination unit for detecting a state change in the state function probability distribution and determining an endpoint. Here, the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability.

An endpoint detection method according to another aspect of the present invention comprises the steps of constructing a hidden Markov model; Obtaining process OES data; And applying the obtained process OES data to the constructed hidden Markov model to detect an endpoint. Building the hidden Markov model comprises obtaining reference OES data using SNR; Normalizing the reference OES data; PCA (Principal Component Analysis) the normalized data to obtain eigenvectors; Generating a reference product by computing a representative loading vector of the reference OES data and the eigenvectors; And acquiring a first segment value corresponding to a first state and a second segment value corresponding to a second state through the generated reference product.

Plasma process diagnostic method according to another aspect of the present invention comprises the steps of building a hidden Markov model; Obtaining OES data; And diagnosing abnormality of the plasma process by applying the obtained OES data to the constructed hidden Markov model. Building the hidden Markov model comprises obtaining reference OES data using SNR; Normalizing the reference OES data; PCA (Principal Component Analysis) the normalized data to obtain eigenvectors; Generating a reference product by computing a representative loading vector of the reference OES data and the eigenvectors; And acquiring a first segment value corresponding to a first state and a second segment value corresponding to a second state through the generated reference product.

Plasma process diagnostic system and an endpoint detection method and apparatus according to the present invention detects the end point by using a spectroscopic analysis method, hidden Markov model and Viterbi algorithm, so even if the semiconductor is miniaturized, whether there is an abnormality and the end point of the plasma process, etc. The advantage is that the same process conditions can be detected accurately.

In particular, since the plasma process diagnosis system detects an end point and the like using the maximum likelihood state sequence tracking method, the end point can be accurately detected even if noise or the like is added to the OES data.

1 is a diagram illustrating a plasma process diagnosis system according to an embodiment of the present invention.
2 is a view illustrating characteristics of light generated in an etching process.
3 is a flowchart illustrating an endpoint detection method according to an embodiment of the present invention.
4 is a diagram illustrating an example of reference OES data.
5 and 6 illustrate a normalization process and a PCA process according to an embodiment of the present invention.
7 illustrates a linear regression model according to an embodiment of the present invention.
8 is a diagram illustrating process products according to an embodiment of the present invention.
9 is a diagram illustrating a process of converting a process product into a state function probability according to an embodiment of the present invention.
10 illustrates state function probability distributions according to an embodiment of the present invention.
11 is a diagram illustrating an endpoint detection process according to an embodiment of the present invention.
12 is a diagram illustrating an end point detection result according to an embodiment of the present invention.
13 is a block diagram illustrating a structure of an endpoint detection unit according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a view showing a plasma process diagnostic system according to an embodiment of the present invention, Figure 2 is a view showing the characteristics of the light generated in the etching process.

The plasma process diagnosis system is a system for diagnosing a process state and abnormality of a plasma process, for example, detecting an end point of an etching process. Here, the plasma process is a process performed by using a plasma, and includes a deposition process, an etching process, and the like.

Hereinafter, the operation of the plasma process diagnosis system will be described in detail by taking an example of detecting the end point in the etching process.

Referring to FIG. 1, the plasma process diagnosis system according to the present embodiment includes a detector 110 and an endpoint detector 112.

In the etching process, a layer made of SiO ₂ or the like is formed on the wafer 104 arranged on the holder 102 in the chamber 100, and a positive photoresist or negative photo is formed on the layer. Negative Photoresist may be formed.

Subsequently, the photoresist is changed into a predetermined pattern using a specific mask, and as a result, a photoresist layer having a predetermined pattern is formed on the layer.

Subsequently, the layer is etched using a plasma in a pattern corresponding to the pattern of the photoresist. Here, the plasma may be supplied to the chamber 100 from the outside, and the gas is changed into plasma by applying a predetermined voltage to the chamber 100 while a specific gas such as CF ₄ is supplied to the chamber 100. May be

The layer reacts with the plasma, in which light is emitted. Here, the light may have various wavelengths according to the by-product after the reaction as shown in FIG. 2 (A). However, in the end point detection method of the present invention, as described later, the end point was detected using light of various wavelength bands including the wavelength of CO.

The detector 110 detects light generated in the reaction process, and may be, for example, an optical emission spectrometer (OES) sensor.

According to an embodiment of the present invention, the detector 110 converts the detected light into OES data. Here, the OES data may be represented as an M × N matrix, where M and N are integers. However, the installation position and structure of the sensing unit 110 is not particularly limited as long as the sensing unit 110 can sense the light.

An end point detector 112 detects an end point of the etching process using the OES data. Specifically, the intensity of light having a specific wavelength is changed with time as shown in FIG. 2 (B). In this case, the endpoint detecting unit 112 determines the point where the state is changed to an end point (about 99 in FIG. 2). Judging by second).

According to one embodiment of the invention, the endpoint detection unit 112 is principal component analysis (Principal Component Analysis, PCA) and Hidden Markov Model, in particular Semi-hidden markov model and segment The endpoint may be detected using a model in which a hidden hidden markov model is mixed. Detailed description thereof will be described later with reference to the accompanying drawings.

In other words, the plasma process diagnosis system of the present embodiment can detect process conditions such as endpoint detection and abnormality of processes using light emitted as a result of reaction in an etching process or the like.

Hereinafter, the endpoint detection process will be described in detail with reference to the accompanying drawings. However, the overall process for the endpoint detection process will be described schematically, and then the detailed process for each step will be described.

First, the entire process of the endpoint detection process will be described.

3 is a flowchart illustrating an endpoint detection method according to an embodiment of the present invention, and FIG. 4 is a diagram illustrating an example of reference OES data. 5 and 6 illustrate a normalization process and a PCA process according to an embodiment of the present invention, and FIG. 7 illustrates a linear regression model according to an embodiment of the present invention.

Referring to FIG. 3, the endpoint detecting unit 112 selects a part of light acquired in an etching process using SNR, for example, to obtain reference OES data and process OES data (S300).

Subsequently, the endpoint detection unit 112 normalizes the obtained reference OES data (S302). Here, the reference OES data is data obtained by extracting light having an SNR greater than or equal to a predetermined value among the light detected by the detector 110, and may be read and normalized according to a request in a state previously stored in a memory.

For example, the reference OES data may be represented as an M (integer) × N (integer) matrix as shown in FIG. 4 and normalized as shown in FIG. 5. Here, M may mean a sample for each wavelength, and N may represent a wavelength sampled at each time.

Subsequently, the endpoint detection unit 112 PCA-processes the normalized reference OES data to obtain an eigenvector (Eigenvector) as shown in FIG. 5 (S304).

Subsequently, the endpoint detection unit 112 generates the reference product by calculating the obtained eigenvector and the reference OES data (S306).

Subsequently, the endpoint detection unit 112 constructs a linear regression model using the generated reference product (S308). In particular, the endpoint detecting unit 112 obtains the first segment value θ ₁ and the second segment value θ ₂ , which are the criteria for state determination.

Subsequently, the endpoint detection unit 112 generates the process product by calculating the process OES data and the eigenvector (S310).

Subsequently, the endpoint detection unit 112 generates a state function probability distribution by converting the generated process product values into state function probabilities (S312).

Subsequently, the endpoint detection unit 112 detects a state transition point through the state function probability distribution, and determines the detected state transition point as an end point (S314). According to an embodiment of the present invention, the endpoint detection unit 112 may use a Viterbi algorithm in the endpoint detection process (S314).

In other words, the endpoint detection unit 112 constructs a linear regression model using reference OES data corresponding to light in a specific wavelength range among the light emitted as a result of the reaction in the etching process, and the constructed linear regression model and the process OES data. The end point is detected using, and a model for detecting the end point, for example, a hidden Markov model, is constructed.

The following describes the detailed process of each step.

First, let's look at the process of selecting OES data.

In the etching process, light generated by etching by-products and reactant gases or light generated by decomposed atoms or molecules (wavelength of etchant, echant) are generated.

Since the light generated by the etching by-products reacts with the reaction gas, it decreases as the etching process proceeds. As a result, the wavelength of light corresponding to the etching by-products decreases as the process proceeds.

On the other hand, the etchant consumes the reaction gas when the etching process proceeds, so that the intensity of the light decreases during the etching process, but when the reaction no longer occurs, the peak of the wavelength of the light associated with the reaction gas increases again.

Therefore, since the light by the etching by-product and the light by the etchant have the opposite characteristics, the subsequent PCA is applied using only one light. In this case, an SNR such as Equation 1 below is used to select the light.

Here, A is data before the end point, B is data after the end point, and S is the standard deviation of all data. For example, if the endpoint is 90 seconds, A may be between 1 and 30 seconds, and B may be between 100 and 130 seconds.

When the SNR is obtained as described above, a graph as shown in FIG. 4A may be obtained. Here, the values above 0 are data by etching by-products, and the values below 0 are data by etchant.

According to an embodiment of the present invention, the endpoint detection method of the present embodiment selects only the wavelength above the reference value as the OES data in the SNR graph. For example, the endpoint detection method may select data having an SNR of 2.3 or more as OES data.

Next, the PCA processing process S302 will be described. Here, PCA (Principal Component Analysis) means a method of eigenvector decomposition of a covariance matrix of a process variable matrix.

Referring to FIG. 5, the endpoint detection method includes a covariance determinant (cov (X)) of Equation 2 below with respect to the normalized reference OES data (expressed as an M × N matrix having M samples and N variables). Calculate the covariance matrix by applying Here, the reference OES data is composed of N pieces of variable p, that is, X ₁₁ , X ₂₂ ,..., X _{p n} . That is, the data may be represented by N points in the p-dimensional space.

Here, X ^T means the transposition matrix of X.

Subsequently, the endpoint detection method calculates a data matrix D by applying Equation 3 below to the covariance matrix as the PCA process.

Subsequently, the endpoint detection method calculates an eigenvalue and an eigenvector by solving the eigenvector problem for the data matrix D as shown in FIG. 6 as the PCA process.

According to an embodiment of the present invention, the endpoint detection method detects an endpoint using the eigenvector.

Next, the process of generating the reference product and the process product will be described.

The reference product includes values of the reference OES data and loading vectors of the eigenvectors AR1 (1) to AR1 (N), AR2 (1) to AR2 (N), ..., ARN (1) to ARN ( N)) is generated by calculating Equation 4 below.

In a practical experiment, four OES data were used to extract respective loading vectors, and a representative product of the extracted loading vectors was calculated with four OES data to generate a reference product.

According to another embodiment of the present invention, a random one of four OES data is selected, a loading vector is extracted from the selected OES data, and the reference product is calculated by calculating the extracted loading vector and the selected OES data. You can also create

The process product is similar to the reference product generation process, and the values of the process OES data and the loading vectors of the eigenvectors AR1 (1) to AR1 (N), AR2 (1) to AR2 (N), ... , ARN (1) to ARN (N)).

In the actual experiment, two OES data were used, and a process product was generated by calculating the representative loading vector and two OES data extracted in the reference product process.

Next, we will look at the process of building the linear regression model. However, before examining the linear regression model, the specificity of end point detection will be described.

As shown in FIG. 2B, the light detected by the sensing unit 110 changes with time, and is clearly distinguished based on the end point. That is, the light may be separated into a first state before the end point and a second state after the end point.

Also, the light always changes from the first state to the second state. Accordingly, the endpoint detection method of the present invention is to construct a linear regression model using a hidden Markov model, for example, set the initial state distribution as shown in Equation 5 below and the transition matrix as shown in Equation 5 below. Set as shown in 6. That is, the initial state distribution is set to be always in the first state, and the light is set to change from the first state to the second state based on the end point.

The linear regression model for the K states is shown in Equation 7 below.

는 회귀 함수를 의미한다. Where 1 ≦ _i ≦ K, e _t is noise,

Is a regression function.

As can be seen in Equation 7, K states can be represented as θ _i , and the values y _t of the process product are expressed as a function of θ _i with noise e _t taken into account. Where noise e _t is mean 0 and variance is σ ² It may be a Gaussian function.

However, since only two states exist in the endpoint detection process, the linear regression model is constructed with a first segmental value θ ₁ and a second segment value θ ₂ as shown in FIG. 8. Can be. Here, the segment values θ ₁ and θ ₂ are values used as criteria for determining the state, and may have a linear characteristic and have a predetermined slope.

Specifically, if the value of the process product is close to θ ₁ , the value is determined to belong to the first state, and if the value of the process product is close to θ ₂ , the value is to belong to the second state. I think the probability is high.

Next, we will look at the process of calculating the state function probability.

8 is a diagram illustrating process products according to an embodiment of the present invention, and FIG. 9 is a diagram illustrating a process of converting process products into state function probabilities according to an embodiment of the present invention. 10 is a diagram illustrating state function probability distributions according to an embodiment of the present invention, and FIG. 11 is a diagram illustrating an endpoint detection process according to an embodiment of the present invention.

8 to 10, the values y _t of the process product are converted into state function probabilities indicating the probability of belonging to the first state and the probability of belonging to the second state through Equation 8 below.

Here, μ _c is a predicted end point known in advance, and may be, for example, a time elapsed 99 seconds from the start of the etching process.

As a result, the values y _{t of} the process product are converted into state function probabilities, so that the process product is represented by one state function probability distribution.

In an actual experiment, four process products as shown in FIG. 8 were transformed through a transformation process as shown in FIG. 9 to obtain state function probability distributions as shown in FIG. 10. Detailed description thereof will be described later.

When the μ _c is set as above, the end point is expected to occur at approximately μ _c ± 20%. As a result, the state sustained distribution of the first state may be expressed as a normal distribution as shown in Equation 9 below. In addition, the state sustain distribution of the second state may be expressed by Equation 10.

이다. here,

to be.

Referring to Equations 9 and 10, the first state corresponding to the process before the etching process is finished may continue as a normal distribution on the basis of the end point, and after the etching process is finished, the first state may be maintained. Since there is no change from two states to another, the probability can be expressed as almost one.

Subsequently, the process of detecting the end point in the state function probability distribution will be described.

11 is a diagram illustrating an endpoint detection process according to an embodiment of the present invention.

As shown in Fig. 11, the endpoint detection method of this embodiment uses a most likely state sequence tracking method. Specifically, the state change of the data sequences (y, probability values in the state function probability distribution) is continuously tracked, but the endpoint is determined by whether the same state is continuously maintained.

For example, sequences before an endpoint are more likely to belong to the first state than to belong to the second state (having the first state), but in some sequences, they may be more likely to belong to the second state. (Having said second state). In this case, since sequences after the sequence having the second state have the first state, that is, the second state is not continuously maintained, the sequence changed from the first state to the second state is determined as an end point. I never do that.

However, if the second states are maintained after the first state is changed to the second state in a particular sequence, the end point is determined as the time point when the first state is changed from the first state to the second state.

For example, as shown in Figs. 10A to 10D, most of the sequences before the endpoint have the first state but some have the second state (i.e. belong to the second state). The probability is higher than the probability of belonging to the first state. In this case, since the sequences after the sequence have the first state, the sequence is determined to be the sequence before the slave CNF exit point.

On the other hand, after the end point (approximately 99 seconds), the majority of sequences have the second state, that is, the second state is continuously maintained, so that the point of time when the first state changes from the first state to the second state is the end point. You can check it.

According to an embodiment of the present invention, the maximum likelihood state sequence tracking method may be performed by using a Viterbi algorithm such as Equation 11 below.

Here, 1 <i <N, and A _ji means the probability of changing from the j state to the i state.

Referring to Equation 11, V ₁ and V ₂ are generated as shown in Equation 12 below in each second, and the final path is determined as the path having the maximum probability as shown in FIG.

By tracking the path through the methods of Equations 11 and 12, it can be seen that in about 99 seconds the second state is maintained after the state of the sequence is changed from the first state to the second state. That is, it can be confirmed that the end point is about 99 seconds.

That is, the endpoint tracking method of the present embodiment detects the endpoint using the maximum likelihood state sequence tracking method, particularly the Viterbi algorithm.

Below, we will look at the actual hidden Markov model construction process for the endpoint detection.

12 is a diagram illustrating an end point detection result according to an embodiment of the present invention.

In the experiment, a hidden Markov model was constructed using the main wavelength band of the light emitted in the wafer etching process. Specifically, a hidden Markov model was constructed by analyzing the 139 values of the principal wavelength band. Here, the slopes of the first segment value θ ₁ and the slopes of the second segment value θ ₂ were determined to be 0.01285 and −0.02, respectively, as shown in FIG. 9.

Subsequently, an end point was detected through process products corresponding to process OES data as shown in FIG. 8, and a hidden Markov model for end point detection was constructed.

Subsequently, as a result of applying to the constructed hidden Markov model using two process products having the same wavelength band as the reference product, the end point was detected as about 99 seconds as shown in FIG. 12.

In other words, it could be confirmed that the constructed hidden Markov model accurately detects the end point. In particular, even though noise was added during OES data acquisition, the endpoint was accurately detected. This is because the path tracking method by the Viterbi algorithm is used.

13 is a block diagram illustrating a structure of an endpoint detection unit according to an embodiment of the present invention.

Referring to FIG. 13, the endpoint detecting unit 112 of the present embodiment includes a control unit 1300, a data selecting unit 1302, a normalization unit 1304, a PCA processing unit 1306, a product generating unit 1308, and a regression model unit ( 1310, a state function probability unit 1312, an endpoint determining unit 1314, and a storage unit 1316.

The data selector 1302 selects OES data corresponding to some of the light generated in the etching process by using the SNR. Here, the selection process may be automatic.

The normalization unit 1304 normalizes the reference OES data.

The PCA processing unit 1306 PCAs the normalized reference OES data.

A product generation unit 1308 generates a reference product by calculating the eigenvector obtained by the reference OES data and the PCA process, and generates a process product by calculating the process OES data and the eigenvector.

The regression model unit 1310 constructs a linear regression model using the reference product, and specifically obtains a first segment value θ ₁ and a second segment value θ ₂ .

The state function probability unit 1312 converts the values of the process product into state function probabilities to generate a state function probability distribution.

An end point determiner 1314 detects an end point through the state function probability distribution, and in particular, detects the end point using a Viterbi algorithm.

The storage unit 1316 stores existing OES data, an end point, and the like.

The controller 1300 generally controls the operations of the components of the endpoint detection unit 118.

The embodiments of the present invention described above are disclosed for purposes of illustration, and those skilled in the art having ordinary knowledge of the present invention may make various modifications, changes, and additions within the spirit and scope of the present invention. Should be considered to be within the scope of the following claims.

100 chamber 102 holder
104: wafer 110: detector
112: endpoint detection unit 1300: control unit
1302: data selection unit 1304: normalization unit
1306: PCA processing unit 1308: product generation unit
1310: regression model unit 1312: state function probability unit
1314: endpoint determination unit 1316: storage unit

Claims (16)

Setting a first segment value corresponding to the first state and a second segment value corresponding to the second state;
Converting process OES data into a state function probability distribution based on the first segment value and the second segment value; And
Detecting an endpoint using the maximum likelihood state sequence tracking method in the state function probability distribution;
And the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability. The method of claim 1, wherein the setting of the first segment value and the second segment value comprises:
Obtaining reference OES data using SNR;
Normalizing the reference OES data;
PCA (Principal Component Analysis) the normalized data to obtain eigenvectors;
Generating a reference product by computing a representative eigenvector among the reference OES data and the eigenvectors; And
Obtaining the first segment value and the second segment value through the generated reference product,
The SNR is,

,
A is data before the expected end point, B is data after the expected end point, and S is a standard deviation of all data. The method of claim 2, wherein the converting into the state function probability distribution comprises:
Generating a process product by computing the process OES data and the representative eigenvector; And
And obtaining the state function probability distribution by converting values of the generated process product into probability values using previously stored endpoints. The method of claim 1, wherein the detecting of the end point comprises:
Detecting whether probability values for each time zone of the state function probability distribution correspond to the first state or the second state; And
Determining the switching time point as the end point when the second state is maintained for a predetermined number of times after being switched from the first state to the second state as a result of tracking the states according to the detection result. End point detection method. The endpoint detection method of claim 4, wherein a Viterbi algorithm is used to track states according to a detection result of whether the first state corresponds to the first state or the second state. The endpoint detection method of claim 1, wherein the endpoint detection method is used to detect an endpoint of an etching process using plasma. A regression model unit for constructing a regression model to obtain a first segment value corresponding to the first state and a second segment value corresponding to the second state;
A state function probability unit for converting process OES data into a state function probability distribution based on the first segment value and the second segment value; And
An endpoint determination unit for determining a termination point by detecting a state change in the state function probability distribution,
And the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability. The method of claim 7, wherein the endpoint detection device,
Obtaining reference OES data using SNR;
A normalizer for normalizing the reference OES data;
A PCA processor configured to obtain eigenvectors by PCA (Principal Component Analysis) the normalized data; And
Further comprising a product generation unit for generating a reference product by calculating a representative eigenvector of the reference OES data and the eigenvectors,
The SNR is,

,
A is data before the expected end point, B is data after the expected end point, S is the standard deviation of the entire data, and the first segment value and the second segment value are obtained through the generated reference product. End point detection device characterized in that. The method of claim 8, wherein the product generation unit generates a process product by calculating process OES data and the representative eigenvector,
The state function probability unit converts the values of the generated process product into probability values using prestored endpoints,
The endpoint determining unit detects whether the probability values for each time zone correspond to the first state or the second state, and after switching states according to the detection result, switch from the first state to the second state. And when the second state is maintained more than a predetermined number of times, determining the switching time point as the end point. The apparatus of claim 9, wherein a Viterbi algorithm is used to track the states according to the detection result, and the endpoint detection device is used to detect an endpoint of an etching process using plasma. Building a hidden Markov model;
Obtaining process OES data; And
Detecting the endpoint by applying the obtained process OES data to the constructed hidden Markov model,
Building the hidden Markov model,
Obtaining reference OES data using SNR;
Normalizing the reference OES data;
PCA (Principal Component Analysis) the normalized data to obtain eigenvectors;
Generating a reference product by computing a representative loading vector of the reference OES data and the eigenvectors; And
And obtaining a first segment value corresponding to a first state and a second segment value corresponding to a second state through the generated reference product. The method of claim 11, wherein the SNR,

,
A is data before the expected end point, B is data after the expected end point, and S is a standard deviation of all data. The method of claim 11, wherein the detecting of the end point comprises:
Converting the process OES data into a state function probability distribution based on the first segment value and the second segment value; And
Detecting an endpoint using the maximum likelihood state sequence tracking method in the state function probability distribution;
And the state function probability distribution is a distribution indicating whether the values of the process OES data belong to the first state or the second state as probability. The method of claim 13, wherein a Viterbi algorithm is used to detect an endpoint using the maximum likelihood state sequence, and the endpoint detection method is used to detect an endpoint of an etching process using plasma. . 12. The method of claim 11, wherein the hidden Markov model is a model in which a semi hidden Markov model and a segment hidden Markov model are used together. Building a hidden Markov model;
Obtaining OES data; And
Applying the obtained OES data to the constructed hidden Markov model to diagnose the abnormality of the plasma process,
Building the hidden Markov model,
Obtaining reference OES data using SNR;
Normalizing the reference OES data;
PCA (Principal Component Analysis) the normalized data to obtain eigenvectors;
Generating a reference product by computing a representative loading vector of the reference OES data and the eigenvectors; And
And obtaining a first segment value corresponding to a first state and a second segment value corresponding to a second state through the generated reference product.

Images