Deep learning-based analysis system and operating method thereof

1 . A semiconductor manufacturing apparatus including a deep learning-based analysis system comprising:

semiconductor equipment to perform a semiconductor process to generate a plurality of products; and

the deep learning-based analysis system comprising:

a detection device including a plurality of sensors to collect data of the semiconductor process and provide corresponding multivariate time-series data; and

an analysis device including at least one processor, the at least one processor being configured to receive the multivariate time-series data and to:

obtain a corresponding correlation degree between corresponding ones of the plurality of sensors based on a first learning model using the received multivariate time-series data as input;

calculate an error score for each sensor based on a second learning model using time-series data for each sensor extracted from the received multivariate time-series data as input;

determine that an error score of a first sensor of the plurality of sensors exceeds a threshold value; and

in response to the determination of the error score of the first sensor exceeding the threshold value, output a control signal or process operation data corresponding to the first sensor through an interface for controlling the equipment process.

2 . The semiconductor manufacturing apparatus of claim 1 , wherein the at least one processor is further configured to generate the first learning model that learns the corresponding correlation degree between corresponding ones of the plurality of sensors based on the multivariate time-series data provided by the plurality of sensors.

3 . The semiconductor manufacturing apparatus of claim 1 , wherein the at least one processor is further configured to generate the first learning model that learns the corresponding correlation degree between corresponding ones of the plurality of sensors based on similarity-related feature characteristics for the multivariate time-series data provided by the plurality of sensors.

4 . The semiconductor manufacturing apparatus of claim 1 , wherein the at least one processor is further configured to:

convert the time-series data for each sensor provided by the plurality of sensors into converted image data for each sensor; and

use the second learning model to detect an abnormality of the semiconductor process based on the converted image data for each sensor and the correlation degree between the plurality of sensors.

5 . The semiconductor manufacturing apparatus of claim 4 , wherein the converted image data for each sensor has a preset data size.

6 . The semiconductor manufacturing apparatus of claim 4 , wherein, for sensors having a corresponding correlation degree greater than or equal to a preset value, the at least one processor is further configured to generate the second learning model by comparing a plurality of corresponding converted image data.

7 . The semiconductor manufacturing apparatus of claim 4 , wherein the at least one processor is further configured to:

calculate reconstruction loss for the converted image data for each sensor of the plurality of sensors based on the second learning model; and

calculate the error score for each sensor based on the calculated reconstruction loss.

8 . A method of manufacturing a semiconductor product with semiconductor equipment having a plurality of sensors, the method comprising:

with the semiconductor equipment, performing a semiconductor process to generate a plurality of semiconductor products;

receiving multivariate time-series data provided by the plurality of sensors of the semiconductor equipment;

obtaining a corresponding correlation degree between corresponding ones of the plurality of sensors based on a first learning model that uses the received multivariate time-series data as input;

calculating an error score for each sensor based on a second learning model using time-series data for each sensor extracted from the received multivariate time-series data as input;

determining that an error score of a first sensor of the plurality of sensors exceeds a threshold value; and

in response to the determination, outputting a control signal or process operation data for controlling the semiconductor process corresponding to the first sensor.

9 . The method of claim 8 , further comprising generating the first learning model that learns the corresponding correlation degree between corresponding ones of the plurality of sensors based on the multivariate time-series data provided by the plurality of sensors.

10 . The method of claim 8 , further comprising generating the first learning model that learns the corresponding correlation degree between corresponding ones of the plurality of sensors based on similarity-related feature characteristics for the multivariate time-series data provided by the plurality of sensors.

11 . The method of claim 8 , further comprising:

converting the time-series data for each sensor provided by the plurality of sensors into converted image data for each sensor; and

using the second learning model to detect an abnormality of the semiconductor process based on the converted image data for each sensor and the correlation degree between the plurality of sensors.

12 . The method of claim 11 , wherein the converted image data for each sensor has a preset data size.

13 . The method of claim 11 , further comprising, for sensors having a corresponding correlation degree greater than or equal to a preset value, generating the second learning model by comparing a plurality of corresponding converted image data.

14 . The method of claim 11 , further comprising:

calculating reconstruction loss for the converted image data for each sensor of the plurality of sensors based on the second learning model; and

calculating the error score for each sensor based on the calculated reconstruction loss.

15 . A deep learning-based analysis device for detecting defects in a semiconductor product processed by semiconductor equipment, the deep learning-based analysis device comprising:

a communication unit configured to receive data through communication with the semiconductor equipment; and

at least one processor,

wherein the at least one processor is configured to:

receive multivariate time-series data provided by a plurality of sensors of the semiconductor equipment through the communication unit;

acquire a corresponding correlation degree between corresponding ones of the plurality of sensors based on a first learning model using the received multivariate time-series data as input;

calculate an error score for each sensor based on a second learning model using the correlation degree between the plurality of sensors as input;

determine that an error score of a first sensor of the plurality of sensors exceeds a threshold value; and

based on the determination of the error score of the first sensor exceeding the threshold value, output a control signal or process operation data for controlling the semiconductor process corresponding to the first sensor through an interface.

16 . The deep learning-based analysis device of claim 15 , wherein the at least one processor is further configured to generate the first learning model that learns the corresponding correlation degree between corresponding ones of the plurality of sensors based on the multivariate time-series data provided by the plurality of sensors.

17 . The deep learning-based analysis device of claim 15 , wherein the at least one processor is further configured to:

convert the time-series data for each sensor provided by the plurality of sensors into converted image data for each sensor; and

use the second learning model to detect an abnormality of a process of the semiconductor equipment based on the converted image data for each sensor and the correlation degree between the plurality of sensors.

18 . The deep learning-based analysis device of claim 17 , wherein the at least one processor is further configured to:

calculate reconstruction loss for the converted image data for each sensor of the plurality of sensors based on the second learning model; and

calculate the error score for each sensor based on the calculated reconstruction loss.