KR20230090501A - Device and method for detecting error of semiconductor etching equipment and computer readable program for the same - Google Patents

Abstract

An apparatus and method for detecting a failure of semiconductor etching equipment and a computer readable program therefor are disclosed. An apparatus for detecting failure of semiconductor etching equipment according to the present invention is a device for detecting failure of semiconductor etching equipment for etching a wafer, and collects time-series spectrum data from an optical emission spectroscopy mounted on the semiconductor etching equipment. A data collection unit to detect an etch endpoint by learning the preprocessed data based on a Time Series Anomaly Detection using Generative Adversarial Networks (TadGAN) algorithm, a preprocessor to extract data at a specific wavelength from among the time series spectrum data, and an etch endpoint detection to detect an etch endpoint. and a failure determination unit that compares the detected etching end point with Ground Truth, which is an actual end point, and determines a failure of the semiconductor etching equipment when an error is greater than a threshold value. Therefore, based on the TadGAN algorithm, by providing a technology for detecting failures of semiconductor etching equipment through unsupervised time series anomaly detection, it is possible to detect failures of semiconductor etching equipment with only a small amount of data and prevent damages caused by failures through preliminary maintenance. It offers the advantage of minimizing

Description

Device and method for detecting error of semiconductor etching equipment and computer readable program for the same

The present invention relates to a device and method for detecting a failure of semiconductor etching equipment and a computer readable program therefor, and more particularly, to a device and method for detecting a failure of semiconductor etching equipment for detecting a failure of semiconductor etching equipment. and a computer readable program therefor.

In general, dry etching using plasma is a very important process in semiconductor manufacturing. When plasma is formed during the etching process, the particles constituting the plasma more activate the chemical reaction. However, the thickness of the material layer to be etched formed on the wafer is not always constant over the entire wafer due to a step or the like, and the etching operation is not uniformly performed over the entire surface of the wafer. Therefore, precise process control is required to etch a desired material layer at a desired portion. Moreover, since the etching material has a selectivity in plasma etching, if the etching proceeds for a long time, a predetermined etching layer is etched and even an unwanted lower layer is etched. In addition, when an etching material having a low selectivity is used to increase etching efficiency, this phenomenon may be further intensified.

Therefore, in order to minimize this problem, a method of changing etching conditions by finding an end-point of the process is widely used in plasma etching, in addition to a method of performing the process for a certain period of time. That is, before a certain point in time, the process is performed by selecting a condition capable of increasing the etching rate, and after a certain point in time, the process is performed by selecting a condition in which the etching rate is slow but the selectivity with the lower layer is high. At this time, a specific point in time of the process means a point in time at which the film quality under the material layer to be etched is revealed, and finding this point in time is called end point detection. A process up to this point is called a main etch, and an etch process after this point is called an over etch.

In recent semiconductor manufacturing processes, quality control, process monitoring, and maintenance are becoming more important as advanced manufacturing technologies become more complex, faster, and automated.

Therefore, there is a need for a system that enables preliminary maintenance by detecting whether or not there is a precise control and failure of the semiconductor etching equipment made by dividing the steps.

Korean Patent Publication No. 2001-0084020

According to one aspect of the present invention, time-series spectrum data is collected from Optical Emission Spectroscopy (OES), data at a specific wavelength is extracted and pre-processed, and TadGAN (Time Series Anomaly Detection using Generative Adversarial Networks) Apparatus, method and device for detecting failure of semiconductor etching equipment, which detects an etching endpoint by learning based on an algorithm, and determines failure of the corresponding semiconductor etching equipment when the error between the detected etching endpoint and the actual endpoint is large, and computer readable for the same Its purpose is to provide a program.

An apparatus for detecting a failure of semiconductor etching equipment according to an embodiment of the present invention is a device for detecting a failure of semiconductor etching equipment for etching a wafer, from an optical emission spectroscopy mounted on the semiconductor etching equipment in time series. A data collection unit that collects spectrum data, a pre-processor that extracts data at a specific wavelength from among the time-series spectrum data, and detects an etch end point by learning the pre-processed data based on a TadGAN (Time Series Anomaly Detection using Generative Adversarial Networks) algorithm. and an etch end point detection unit that compares the detected etch end point with Ground Truth and, when the error is greater than a threshold value, a failure determination unit that determines failure of the semiconductor etching equipment.

On the other hand, the etch endpoint detection unit divides the preprocessed data into a sliding window having a preset size and inputs the splitter as training data of the TadGAN algorithm, and predicts generating prediction data by repeatedly training to distinguish between the training data and random data. A data generator, an error value calculator that calculates a reconstruction error value by comparing the training data and prediction data, and calculates an error value based on the calculated reconstruction error value, and an error value calculator that calculates an error value greater than a predetermined threshold value. and an etch end point detector that detects the section in which the abnormality is detected as the etch end point.

In addition, the predictive data generator includes a first classifier that trains to distinguish between the training data and data obtained by converting the random data into a generator, and a second classifier trained to discriminate between the training data and data obtained by converting the random data into an encoder. A separator may be included.

A failure detection method of semiconductor etching equipment according to another embodiment of the present invention is a failure detection method of semiconductor etching equipment in a device for detecting failure of semiconductor etching equipment for etching a wafer, comprising: Collecting time-series spectral data from optical emission spectroscopy, extracting and pre-processing data at a specific wavelength among the time-series spectral data, and applying TadGAN (Time Series Anomaly Detection using Generative Adversarial Networks) algorithm to the pre-processed data. and detecting an etch endpoint by learning based on the basis and comparing the detected etch endpoint with ground truth, and determining failure of the semiconductor etching equipment when an error is greater than a threshold value.

Meanwhile, the step of detecting the etching end point may include dividing the preprocessed data into sliding windows having a preset size and inputting the data as training data of the TadGAN algorithm, repeatedly training to distinguish between the training data and random data, and generating predicted data. generating, comparing the training data and prediction data to calculate a reconstruction error value, calculating an error value based on the calculated reconstruction error value, and if the calculated error value is greater than a preset threshold value, The method may include detecting and detecting a section in which the abnormality is detected as the etching end point.

In addition, generating the prediction data includes training to distinguish between the training data and data obtained by converting the random data into a generator, and training to distinguish between the training data and data obtained by converting the random data into an encoder. can include

In addition, another embodiment of the present invention may include a computer readable program stored in a computer readable recording medium configured to execute a method for detecting a failure of semiconductor etching equipment.

According to the present invention described above, based on the TadGAN algorithm, a technique for detecting failures of semiconductor etching equipment through unsupervised time series anomaly detection is provided, thereby detecting failures of semiconductor etching equipment with only a small amount of data and through preliminary maintenance. It provides the advantage of minimizing damage due to failure.

1 is a diagram illustrating an example of a semiconductor etching system to which a device for detecting a failure of semiconductor etching equipment according to an embodiment of the present invention is applied.
2 is a diagram showing a specific configuration of an apparatus for detecting a failure of semiconductor etching equipment according to an embodiment of the present invention.
FIG. 3 is a diagram showing a specific configuration of an etching endpoint detection unit shown in FIG. 2 .
FIG. 4 is a diagram showing an example of time-series spectrum data collected by the data collection unit shown in FIG. 2 .
FIG. 5 is a diagram illustrating an example of time-series spectrum data collected by the data collection unit shown in FIG. 2 .
FIG. 6 is a view showing a result of detecting an etch end point by the etch end point detector shown in FIG. 2 and ground truth.
7 is a diagram illustrating a method of detecting a failure of semiconductor etching equipment according to another embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The detailed description of the present invention which follows refers to the accompanying drawings which illustrate, by way of illustration, specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable one skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different from each other but are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in another embodiment without departing from the spirit and scope of the invention in connection with one embodiment. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the detailed description set forth below is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is limited only by the appended claims, along with all equivalents as claimed by those claims. Like reference numbers in the drawings indicate the same or similar function throughout the various aspects.

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

1 is a diagram showing an example of a semiconductor etching system to which a device for detecting a failure of semiconductor etching equipment according to an embodiment of the present invention is applied, and FIG. 2 is a diagram showing a failure of semiconductor etching equipment according to an embodiment of the present invention. FIG. 3 is a diagram showing a specific configuration of an etching endpoint detection unit shown in FIG. 2, and FIG. 4 is time-series spectrum data collected by the data collection unit shown in FIG. 2. FIG. 5 is a diagram showing an example of time-series spectrum data collected by the data collection unit shown in FIG. 2, and FIG. 6 is a diagram showing an etching endpoint in the etching endpoint detection unit shown in FIG. 2 It is a drawing showing the result of detecting and the appearance of Ground Truth.

The semiconductor etching system to which the device for detecting a failure of the semiconductor etching equipment according to the present embodiment is applied is the semiconductor etching equipment 10, the optical reflection spectrometer (OES: Optical Emission Spectroscopy) coupled to the semiconductor etching equipment 10, OES equipment ( 20) and a device 100 for detecting failure of semiconductor etching equipment.

The semiconductor etching equipment 10 is equipment for etching a wafer by generating plasma to an etching material in a plasma chamber.

The OES equipment 20 measures and collects the intensity according to the wavelength of all radicals of the plasma in the plasma chamber by dispersing the light into a single light spectrum, and collects the intensity information according to the wavelength of the collected radicals in the time-series spectrum. It is a device that provides data.

The device 100 for detecting a failure of semiconductor etching equipment observes a phenomenon that occurs when plasma is generated in the semiconductor etching equipment 10 to etch the surface of a wafer, and the etching end point so that the wafer is etched to a desired portion. is detected, and based on the detected etching end point, it is determined whether or not the semiconductor etching equipment has failed. To this end, the device 100 for detecting failure of semiconductor etching equipment collects time-series spectrum data from the OES equipment 20 coupled with the semiconductor etching equipment 10 that etches the wafer, and based on this, determines the etching end point and failure. detect

More specifically, the apparatus 100 for detecting a failure of semiconductor etching equipment according to the present embodiment includes a data collection unit 110, a pre-processing unit 121, an etching endpoint detection unit 122, a failure determination unit 123, and A memory unit 130 is included.

The data collection unit 110 collects time-series spectrum data from OES equipment.

Meanwhile, the time-series spectrum data is intensity data for each wavelength over time, where the wavelength may have a value in the range of 200 to 800 nm. Chemical reactions that occur in a chamber when etching occurs in semiconductor etching equipment have a unique light spectrum for each reaction. As shown in FIG. 4, this time-series spectrum data can be expressed as a two-dimensional graph as a change in the intensity value for each wavelength over time, and as shown in FIG. 5, a three-dimensional graph having three axes of time, wavelength, and intensity. can be expressed as

The data collection unit 110 may collect time-series spectrum data measured on a plurality of wafers. For example, the data collection unit 110 may preferably collect time-series spectrum data measured on at least 50 wafers.

As described above, FIGS. 4 and 5 are examples of time-series spectrum data collected from one wafer. Accordingly, the time-series spectrum data collected by the data collection unit 110 has a vast amount of time-dependent intensity information for at least 2048 wavelengths.

Therefore, the pre-processing unit 121 according to the present embodiment extracts data at a specific wavelength from among the time-series spectrum data and reduces the data dimension.

At this time, the pre-processing unit 121 extracts intensity data over time at a specific wavelength that has a large influence on detecting an etching end point. For example, the pre-processing unit 121 may extract intensity data over time at specific wavelengths closely related to the etching endpoint, such as 387 nm, 440 nm, 520 nm, and 777 nm. On the other hand, the specific wavelength is illustrative, and may be selected within a range of wavelengths that are highly related to the wavelength of the intrinsic emission of the material generated in the semiconductor etching process.

The etch endpoint detection unit 122 detects the etch endpoint by learning the preprocessed data based on a Time Series Anomaly Detection using Generative Adversarial Networks (TadGAN) algorithm. Here, TadGAN is an unsupervised anomaly detection algorithm built using a Generative Adversarial Network (GAN).

That is, the etch end point detection unit 122 according to the present embodiment defines an abnormal section through TadGAN as an EPD section and detects an etch end point. More specifically, the etch endpoint detection unit 122 constructs prediction data, which is reconstructed data, based on the training data, and when new data is received, calculates an error value between the previously constructed prediction data and the new data, An abnormal section is detected by comparing the error value with a threshold value.

Hereinafter, a configuration for detecting a specific etch endpoint of the etch endpoint detector 122 will be described.

The etch endpoint detector 122 includes a divider 122-1, a prediction data generator 122-2, an error value calculator 122-3, and an etch endpoint detector 122-4.

First, the divider 122-1 divides the time series data into a plurality of signal segments. To this end, the divider 122-1 according to the present embodiment divides the pre-processed data from the pre-processing unit 121 into sliding windows of a preset size and inputs them as training data.

The prediction data generator 122-2 repeatedly trains to distinguish the training data from the random data to generate prediction data, which is reconstructed data. To this end, the prediction data generator 122-2 includes a first classifier for training to distinguish between the training data and the data obtained by converting the random data into a generator, the training data, and the random data into an encoder (Encoder). ) may include a second discriminator that trains to discriminate data changed to Here, the generator and the first classifier may be composed of a recurrent neural network of LSTM (Long Short-Term Memory) in order to effectively learn the temporal correlation of the time series distribution, and the first classifier and the second classifier effectively convert the time series data To reconstruct, we conduct training to distinguish between training data and random data based on Cycle Consistency Loss.

The error calculator 122-3 calculates a reconstruction error by comparing the training data with the predicted data, and calculates a final error based on the calculated reconstruction error. The error value calculator 122-3 calculates the difference between training data, which is actual data, and predicted data, which is reconstructed data, as a reconstruction error value, and uses the reconstruction error value to calculate training in the first classifier and the second classifier. The final error value can be calculated by adding up one critical score.

The etch endpoint detector 122-4 detects an anomaly when the calculated final error value is greater than a preset threshold value, and detects a section in which the anomaly is detected as the etch endpoint.

The existing CNN algorithm is a supervised learning model, which has the advantage of high accuracy due to human intervention in the target value, but has the disadvantage of requiring a long learning time and a large amount of training data. Therefore, unlike the CNN-based etch endpoint detection, the etch endpoint detection unit according to the present embodiment detects the etch endpoint based on TadGAN, an unsupervised learning method that does not set a target value, and can quickly detect the etch endpoint with only a small amount of data. have an advantage

In addition, as a result of testing the TadGAN-based etch end point detection method according to this embodiment on 50 wafers, the average accuracy for 50 wafers is about reached 98.83%. Therefore, according to the present invention, it is possible to implement a high-speed and high-accuracy etching endpoint detection method even with a small amount of data.

On the other hand, if the error between the etch end point detected by the etch end point detector 122 and the ground truth, which is the actual end point, increases, this may indicate a failure of the semiconductor etching equipment 10 . Therefore, the apparatus 100 for detecting a failure of semiconductor etching equipment according to the present embodiment compares the detected etching endpoint with the actual endpoint, Ground Truth, and detects a failure of the semiconductor etching equipment when the error is greater than a preset threshold. It includes a failure determination unit 123 that determines.

For example, FIG. 6 is a diagram illustrating an etch endpoint detected by the etch endpoint detector 122 through time-series spectrum data for four wavelengths and an actual endpoint, Ground Truth. According to FIG. 6, an error occurs between the detected etching end point and the actual end point. If the error is greater than the threshold value, the failure determination unit 123 determines that the semiconductor etching equipment 10 has failed and allows the user to use the corresponding equipment. It can check for failures and guide you to carry out preventive maintenance.

Finally, the memory unit 130 stores information on a threshold value for detecting an anomaly in the etch end point detector 122-4 and a threshold value for an error between the etch end point and the actual end point in the failure determination unit 123. can

7 is a diagram illustrating a method of detecting a failure of semiconductor etching equipment according to another embodiment of the present invention.

According to FIG. 7 , a method for detecting failure of semiconductor etching equipment according to the present embodiment is a method for detecting failure of semiconductor etching equipment for etching a wafer, and includes an optical reflection spectrometer (OES: Collecting time-series spectrum data from Optical Emission Spectroscopy (S10), extracting and pre-processing data at a specific wavelength among the time-series spectrum data (S20), learning the pre-processed data based on the TadGAN algorithm to obtain an etching endpoint Detecting (S30) and comparing the detected etching end point with Ground Truth, and when the error is greater than a threshold, determining failure of the semiconductor etching equipment (S40).

In addition, the step of detecting the etching end point (S30) is the step of dividing the preprocessed data into sliding windows having a preset size and inputting them as training data of the TadGAN algorithm, repeatedly training to distinguish between the training data and random data, Generating prediction data, calculating a reconstruction error value by comparing the training data and prediction data, and calculating an error value based on the calculated reconstruction error value, and the calculated error value is greater than a preset threshold value. The method may include detecting an anomaly if it is large, and detecting a section in which the anomaly is detected as the etch end point.

In addition, generating the prediction data includes training to distinguish between the training data and data obtained by converting the random data into a generator, and training to distinguish between the training data and data obtained by converting the random data into an encoder. can include

According to the present invention described above, based on the TadGAN algorithm, a technique for detecting failures of semiconductor etching equipment through unsupervised time series anomaly detection is provided, thereby detecting failures of semiconductor etching equipment with only a small amount of data and through preliminary maintenance. It provides the advantage of minimizing damage due to failure. An operation by the etching endpoint detection method according to the embodiments described above may be at least partially implemented as a computer program and recorded on a computer-readable recording medium. A program for implementing an operation by a deep learning-based super-resolution image processing method according to embodiments is recorded and a computer-readable recording medium includes all types of recording devices in which data readable by a computer is stored. include Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. In addition, computer-readable recording media may be distributed in computer systems connected through a network, and computer-readable codes may be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing this embodiment can be easily understood by those skilled in the art to which this embodiment belongs.

Although the above has been described with reference to embodiments, it will be understood that those skilled in the art can variously modify and change the present invention without departing from the spirit and scope of the present invention described in the claims below. You will be able to.

10: semiconductor etching equipment
20: OES equipment
100: Device for detecting failure of semiconductor etching equipment

Claims (7)

A device for detecting failures of semiconductor etching equipment that etches wafers,
a data collection unit for collecting time-series spectrum data from an Optical Emission Spectroscopy (OES) mounted on the semiconductor etching equipment;
a pre-processing unit extracting data at a specific wavelength from among the time-series spectrum data;
an etch endpoint detector configured to detect an etch endpoint of the wafer by learning the preprocessed data based on a Time Series Anomaly Detection using Generative Adversarial Networks (TadGAN) algorithm; and
A device for detecting failure of semiconductor etching equipment comprising a failure determination unit that compares the detected etching endpoint with Ground Truth, which is an actual endpoint, and determines failure of the semiconductor etching equipment when the error is greater than a threshold value. According to claim 1,
The etching endpoint detection unit,
a divider dividing the preprocessed data into sliding windows having a predetermined size and inputting the data as training data of the TadGAN algorithm;
a predictive data generator generating predictive data by performing repeated training to distinguish the training data from the random data;
an error value calculator calculating a reconstruction error value by comparing the training data and prediction data, and calculating an error value based on the calculated reconstruction error value; and
An apparatus for detecting a failure of semiconductor etching equipment comprising an etch endpoint detector that detects an anomaly when the calculated error value is greater than a preset threshold value and detects a section in which the anomaly is detected as the etch endpoint. According to claim 2,
The predictive data generator,
A first classifier for training to distinguish between the training data and the data obtained by converting the random data into a generator;
A device for detecting a failure of semiconductor etching equipment including a second classifier for training to discriminate between the training data and data obtained by converting the random data into an encoder. A method for detecting failure of semiconductor etching equipment in a device for detecting failure of semiconductor etching equipment for etching a wafer,
Collecting time-series spectrum data from an optical emission spectroscopy mounted on the semiconductor etching equipment;
extracting and pre-processing data at a specific wavelength among the time-series spectrum data;
Detecting an etch end point by learning the preprocessed data based on a TadGAN (Time Series Anomaly Detection using Generative Adversarial Networks) algorithm; and
Comparing the detected etching end point with Ground Truth, which is an actual end point, and determining a failure of the semiconductor etching equipment when the error is greater than a threshold value. According to claim 4,
The step of detecting the etch endpoint,
Dividing the preprocessed data into sliding windows of a preset size and inputting the data as training data for the TadGAN algorithm;
Generating predictive data by repeatedly training to distinguish the training data from the random data;
calculating a reconstruction error value by comparing the training data and prediction data, and calculating an error value based on the calculated reconstruction error value; and
and detecting an anomaly when the calculated error value is greater than a preset threshold and detecting a section in which the anomaly is detected as the etching end point. According to claim 5,
Generating the prediction data,
training to distinguish between the training data and data obtained by converting the random data into a generator; and
A method of detecting a failure of semiconductor etching equipment comprising the step of training to distinguish between the training data and data obtained by converting the random data into an encoder. A computer readable program stored on a computer readable recording medium, configured to execute the method for detecting a failure of semiconductor etching equipment according to claims 4 to 6.

Images