Search

KR-102962964-B1 - Data processing system, and semiconductor facility abnormal data detection system

KR102962964B1KR 102962964 B1KR102962964 B1KR 102962964B1KR-102962964-B1

Abstract

The technical concept of the present invention comprises: an input unit that compresses input data through an encoder; an output unit that restores the compressed input data through a decoder and outputs at least one output data; and a detection unit that determines whether the input data is abnormal by comparing the difference between the at least one output data output from the output unit and the input data. The present invention provides a data processing system characterized by including, wherein the output unit comprises a first output unit that outputs first output data through a first decoder, a second output unit that outputs second output data through a second decoder, and a third output unit that outputs third output data through a third decoder, wherein the first output unit receives the input data as an input, the second output unit receives the first output data as an input, and the third output unit receives the second output data as an input, wherein the first decoder restores the input data compressed by the encoder and provides the first output data, the second decoder restores the first output data compressed by the encoder and provides the second output data, and the third decoder restores the second output data compressed by the encoder and provides the third output data.

Inventors

  • 하민성

Assignees

  • 삼성전자주식회사

Dates

Publication Date
20260508
Application Date
20230821

Claims (10)

  1. An input unit that compresses input data through an encoder; An output unit that restores the above-mentioned compressed input data through a decoder and outputs at least one output data; and a detection unit that determines whether the input data is abnormal by comparing the difference between the at least one output data output from the output unit and the input data; The above output unit includes a first output unit that outputs first output data through a first decoder, a second output unit that outputs second output data through a second decoder, and a third output unit that outputs third output data through a third decoder. The first output unit receives the input data as an input, the second output unit receives the first output data as an input, and the third output unit receives the second output data as an input. The first decoder restores the input data compressed by the encoder and provides the first output data, and The second decoder restores the first output data compressed by the encoder and provides the second output data. A data processing system characterized in that the third decoder restores the second output data compressed by the encoder and provides the third output data.
  2. In paragraph 1, A data processing system characterized in that each of the first decoder, the second decoder, and the third decoder is learned through adversarial learning.
  3. In paragraph 1, A data processing system characterized in that each of the encoder, the first decoder, the second decoder, and the third decoder includes an active layer, and at least one of the encoder, the first decoder, the second decoder, and the third decoder includes a sigmoid function as the active layer.
  4. In paragraph 1, The above encoder includes an input layer, a hidden layer, and an active layer, and Each of the above first decoder, second decoder, and third decoder includes an output layer, a hidden layer, and an active layer, and A data processing system characterized in that each of the above encoder, first decoder, second decoder, and third decoder includes three hidden layers.
  5. In paragraph 4, Each of the above first decoder, second decoder, and third decoder further includes a hidden layer and an output layer, and The above active layer is formed between the hidden layers, and between the hidden layer and the output layer, and A data processing system characterized by having two active layers formed between the output layer and the hidden layer.
  6. In paragraph 5, A data processing system characterized in that the active layer closest to the output layer includes a sigmoid function.
  7. A semiconductor equipment sensor unit comprising at least one semiconductor equipment and at least one sensor provided in each of the semiconductor equipment; A data processing unit that processes input data provided from the above sensors to detect abnormal data; and A storage unit that collects the above abnormal data and displays it through a display; The above data processing unit is, An encoder that compresses input data, a first decoder that outputs first output data restored from the compressed input data, A second decoder that outputs second output data restored from first output data compressed by the encoder above, and It includes a third decoder that outputs third output data restored from second output data compressed by the encoder above, and The above first decoder and second decoder are learned through adversarial learning, and A semiconductor equipment abnormal data detection system characterized in that the second decoder and the third decoder are learned through mutual adversarial learning.
  8. In Paragraph 7, The above encoder includes an input layer, a hidden layer, and an active layer, and Each of the above first decoder, second decoder, and third decoder includes an output layer, a hidden layer, and an active layer, and A semiconductor equipment abnormal data detection system characterized in that each of the above encoder, first decoder, second decoder, and third decoder includes three hidden layers.
  9. In paragraph 8, A semiconductor equipment abnormal data detection system characterized in that at least one of the first decoder, the second decoder, and the third decoder comprises a ReLU function and a sigmor function between the output layer and the hidden layer closest to the output layer.
  10. In Paragraph 9, A semiconductor equipment abnormal data detection system characterized in that the active layer closest to the output layer includes a sigmoid function.

Description

Data processing system, and semiconductor facility abnormal data detection system The present invention relates to a data processing system and a semiconductor equipment abnormal data detection system, and more specifically, to a data processing system using an autoencoder and a semiconductor equipment abnormal data detection system. Data processing systems utilizing data collected through sensors employ AI technologies based on supervised or unsupervised learning, depending on whether the collected data is labeled. While supervised learning requires data pre-labeled as normal or abnormal, the frequency of abnormalities in actual industrial settings is very low; therefore, acquiring abnormal data and labeling the collected data incurs time and costs. To overcome the above problems, unsupervised learning-based data processing systems are sometimes used, which proceed with training under the assumption that the collected data is normal. Generally, abnormal data is detected by calculating the difference between the input and output through a process of reducing and restoring the training data. However, unsupervised learning-based systems have the disadvantage that their performance depends on hyperparameters, resulting in unstable accuracy compared to supervised learning-based anomaly detection systems, and that thresholds for detecting outliers must be calculated or arbitrarily specified. FIG. 1 is a schematic diagram illustrating a semiconductor equipment abnormal data detection system according to exemplary embodiments of the present invention. Figure 2 is a schematic diagram showing the semiconductor equipment sensor section of Figure 1. FIG. 3a is a flowchart for explaining the data processing unit of FIG. 1. FIGS. 3B and FIGS. 3C are schematic diagrams for explaining the data processing unit of FIG. 1. Figure 4a is a schematic diagram showing part AA of Figure 3a in a schematic manner. Figures 4b and 4c are graphs schematically showing the Relu function and the sigmoid function. Figure 5 is a diagram illustrating the effect of improving the internal performance of the autoencoder in the data processing unit. Figures 6a and 6b are diagrams illustrating the performance improvement effect according to the increase in the number of autoencoders in the data processing unit. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Identical components in the drawings are denoted by the same reference numerals, and redundant descriptions thereof are omitted. FIG. 1 is a schematic diagram illustrating a semiconductor equipment abnormal data detection system according to exemplary embodiments of the present invention. FIG. 2 is a schematic diagram illustrating a semiconductor equipment sensor unit of FIG. 1. FIG. 3a is a flowchart for explaining a data processing unit of FIG. 1, and FIG. 3b and FIG. 3c are schematic diagrams for explaining a data processing unit of FIG. 1. Referring to FIGS. 1 to 3b, a semiconductor equipment abnormal data detection system (1) may include a semiconductor equipment sensor unit (10), a data processing unit (20), and a storage unit (30). The semiconductor equipment sensor unit (10) may include a semiconductor equipment (11) and a sensor (13) that senses physical quantities of the semiconductor equipment (11). According to exemplary embodiments, a plurality of sensors (13) may be provided in a single semiconductor equipment (11). The plurality of sensors (13) may be configured to measure physical quantities inside the semiconductor equipment (11). For example, the sensor (13) may measure pressure inside the semiconductor equipment (11), the flow rate of fluid flowing into the semiconductor equipment (11), RF (Radio Frequency) power applied inside the semiconductor equipment (11), the temperature inside the semiconductor equipment (11), etc. However, it is not limited thereto, and the sensor (13) is sufficient as a device capable of measuring physical quantities necessary for manufacturing semiconductors. According to exemplary embodiments, a plurality of semiconductor facilities (11) may be provided. According to exemplary embodiments, the plurality of semiconductor facilities (11) may be of the same type of facility or of different types of facility. Additionally, each semiconductor facility (11) may include at least one sensor (13). Ultimately, in the process of manufacturing semiconductors, multiple semiconductor facilities (11) are required, and a large amount of data can be measured through multiple sensors (13) provided to the multiple semiconductor facilities (11). In this case, it may take a long time to collect and analyze the data from all the sensors (13). The data processing unit (20) may include an input unit (20i) and an output unit (20o). The input unit (20i) may include an encoder (21). Input data provided from the semiconductor equipment sensor unit (10) may be input to the input unit (20i). The encode