US-12619224-B2 - Processor including modified radial basis function (RBF) neural network and method of providing the modified RBF neural network

US12619224B2US 12619224 B2US12619224 B2US 12619224B2US-12619224-B2

Abstract

Provided is a method of providing a modified radial basis function (RFB) neural network. The method includes providing the modified RBF neural network configured to determine a breakdown of semiconductor equipment, wherein the modified RBF neural network assigns, to each of components of the measurement data, a standardization coefficient dependent on the components of the measurement data.

Inventors

Jisub LEE

Assignees

SEMES CO., LTD.

Dates

Publication Date: 20260505
Application Date: 20230417
Priority Date: 20220503

Claims (20)

1 . A method of using a system comprising a processor to determine in real time whether semiconductor equipment that is manufacturing semiconductor devices is broken down, the method comprising and by the processor: obtaining a radial basis function (RBF) neural network configured to predict, based on n-dimensional measurement data with respect to the semiconductor equipment, a likelihood of the semiconductor equipment being broken down, wherein n is an integer; modifying the RBF neural network to obtain a modified RBF neural network that assigns, to each of components of the measurement data, a standardization coefficient dependent on the components of the measurement data; and using the modified RBF neural network instead of the RBF neural network to predict, in the real time while the semiconductor equipment is manufacturing the semiconductor devices, the likelihood of the semiconductor equipment being broken down to prevent defects from being caused in the semiconductor devices by the semiconductor equipment actually being broken down.
2 . The method of claim 1 , wherein the standardization coefficient is n-dimensional.
3 . The method of claim 1 , wherein the standardization coefficient is determined based on a standard deviation of a corresponding component of the measurement data.
4 . The method of claim 1 , wherein the standardization coefficient prevents an excessive increase or an excessive decrease of an effect of each of the components of the measurement data on a calculation by the modified RBF neural network.
5 . The method of claim 1 , wherein, when the semiconductor equipment is determined by the processor to be in a breakdown state, an RBF value calculated by the modified RBF neural network is 1.
6 . The method of claim 1 , wherein, when the semiconductor equipment is determined by the processor to be in a normal state, an RBF value calculated by the modified RBF neural network is 0.
7 . The method of claim 1 , wherein the modified RBF neural network is configured to predict the likelihood of the semiconductor equipment being broken down using Equation 1 below: S j = exp ( - ∑ i = 1 n ⁢ w i *  Z i , j - X i  2 2 ⁢ σ i 2 ∑ i = 1 n ⁢ w i ) [ Equation ⁢ 1 ] where S j is a modified RBF value calculated based on the measurement data, w i selects an i th component of the measurement data, σ i standardizes the i th component of the measurement data, Z i,j is an i th component of standard data indicating a normal state of the semiconductor equipment, and X i is the i th component of the measurement data.
8 . The method of claim 7 , wherein w i has a value of 0 or 1.
9 . The method of claim 7 , wherein w i has a value determined based on a correlation coefficient between components of the standard data.
10 . A system comprising: a memory; and a processor coupled to the memory, the system being a computing device and the processor is configured to: generate, in real time, a prediction indicating a likelihood of semiconductor equipment being broken down using a modified radial basis function (RBF) neural network and measurement data with respect to the semiconductor equipment, the semiconductor equipment being used in the real time while the prediction is generated to manufacture semiconductor devices and the modified RBF neural network being based on following Equation 1: S j = exp ( - ∑ i = 1 n ⁢ w i *  Z i , j - X i  2 2 ⁢ σ i 2 ∑ i = 1 n ⁢ w i ) [ Equation ⁢ 1 ] where S j is a modified RBF value calculated based on the measurement data, W i selects an i th component of the measurement data, σ i standardizes the i th component of the measurement data, Z i,j is an i th component of standard data indicating a normal state of the semiconductor equipment, and X i is the i th component of the measurement data; and using the prediction indicating the likelihood of the semiconductor equipment being broken down to prevent defects being caused in the semiconductor devices being manufactured by the semiconductor equipment in the real time.
11 . The system of claim 10 , wherein w i is determined by Matrix C calculated based on the standard data and following Equation 2: C =|corr( Z,Z )| [Equation 2] wherein Matrix C is a correlation coefficient between components of the standard data.
12 . The system of claim 11 , wherein, when an i th component of any one selected from among rows of Matrix C is less than or equal to a threshold value, w i has a value of 0.
13 . The system of claim 11 , wherein, when an i th component of any one selected from among rows of Matrix C is greater than or equal to a threshold value, w i has a value of 1.
14 . The system of claim 13 , wherein, wherein the threshold value is in a range of about 0.5 to about 0.7.
15 . The system of claim 10 , wherein σi is a standard deviation of the i th component of the measurement data.
16 . A method of using a system comprising a processor to determine in real time whether semiconductor equipment that is manufacturing semiconductor devices is broken down, the method comprising and by the processor: obtaining a radial basis function (RBF) neural network configured to predict, based on n-dimensional measurement data with respect to the semiconductor equipment, a likelihood of the semiconductor equipment being broken down, wherein n is an integer and the RBF neural network is based on following equation 1: S j ′ = exp ⁢ ( -  Z j → - X →  2 σ 2 ) [ Equation ⁢ 1 ] where S′ j is an RBF value calculated based on measurement data of the semiconductor equipment, {right arrow over (Z j )} is standard data indicating a normal state of the semiconductor equipment, {right arrow over (X)} is the measurement data, and σ is a standard deviation of the measurement data; modifying the RBF neural network to obtain a modified RBF neural network, the modified neural network being based on following Equation 2: S j = exp ( - ∑ i = 1 n ⁢ w i *  Z i , j - X i  2 2 ⁢ σ i 2 ∑ i = 1 n ⁢ w i ) [ Equation ⁢ 2 ] where S j is a modified RBF value calculated based on the measurement data, w i selects an i th component of the measurement data, σi standardizes the i th component of the measurement data, Z i,j is an i th component of the standard data indicating the normal state of the semiconductor equipment, and X i is the i th component of the measurement data; and using the modified RBF neural network instead of the RBF neural network to predict, in the real time while the semiconductor equipment is manufacturing the semiconductor devices, the likelihood of the semiconductor equipment being broken down to prevent defects from being caused in the semiconductor devices by the semiconductor equipment actually being broken down.
17 . The method of claim 16 , wherein modifying the RBF neural network to obtain the modified RBF neural network comprises: providing estimation values with respect to {right arrow over (Z j )}; and σ; and updating {right arrow over (Z j )} and σ to improve accuracy of the RBF neural network.
18 . The method of claim 17 , wherein the providing of the estimation values with respect to {right arrow over (Z j )} and σ comprises use of unsupervised learning, and the updating of {right arrow over (Z j )} and σ comprises use of supervised learning.
19 . The method of claim 16 , wherein w i is determined by Matrix C calculated based on the standard data and following Equation 3: C =|corr( Z,Z )| [Equation 3] wherein each of components of Matrix C is an absolute value of a correlation coefficient between components of the standard data.
20 . The method of claim 19 , wherein, when an i th component of any one selected from among rows of Matrix C is less than or equal to a threshold value, w i has a value of 0, when the i th component of any one selected from among the rows of Matrix C is greater than or equal to the threshold value, w i has a value of 1, and the threshold value is in a range of about 0.5 to about 0.7.

Description

CROSS-REFERENCE TO RELATED APPLICATION This application is based on and claims priority under 35 U.S.C. § 119 to Korean Patent Application No. 10-2022-0055028, filed on May 3, 2022, in the Korean Intellectual Property Office, the disclosure of which is incorporated by reference herein in its entirety. BACKGROUND 1. Field One or more embodiments relate to a processor including a modified radial basis function (RBF) neural network and a method of providing the modified RBF neural network. 2. Description of Related Art A yield rate of a semiconductor device is directly related to manufacturing costs, therefore the yield rate is the most essential element in semiconductor device manufacturing. To improve the yield rate of a semiconductor device, it is highly important to monitor a state of semiconductor equipment in real time and predict a breakdown of the semiconductor equipment. A breakdown of semiconductor equipment causes defects in a semiconductor device, and in certain cases, induces immense repair costs. To prevent these problems, a method and a system for predicting the breakdown of semiconductor equipment are required. SUMMARY One or more embodiments include a processor including a modified radial basis function (RBF) neural network and a method of providing the modified RBF neural network. Objectives of the disclosure are not limited to those mentioned above, and other unmentioned objectives will be clearly understood by one of ordinary skill in the art from the descriptions below. Additional aspects will be set forth in part in the description which follows and, in part, will be apparent from the description, or may be learned by practice of the presented embodiments of the disclosure. According to one or more embodiments, there is provided a method of providing a modified radial basis function (RBF) neural network. The method includes: providing an RBF neural network configured to determine, based on n-dimensional measurement data with respect to semiconductor equipment, a breakdown of the semiconductor equipment, wherein n is an integer; and based on the RBF neural network, providing a modified RBF neural network, wherein the modified RBF neural network assigns, to each of components of the measurement data, a standardization coefficient dependent on the components of the measurement data. The standardization coefficient may be n-dimensional. The standardization coefficient may be determined based on a standard deviation of a corresponding component of the measurement data. The standardization coefficient may prevent an excessive increase or an excessive decrease of an effect of each of the components of the measurement data on a calculation of the RBF neural network. When the semiconductor equipment has a breakdown, an RBF value calculated by the modified RBF neural network may be 1. When the semiconductor equipment has a breakdown, an RBF value calculated by the modified RBF neural network may be 0. The modified RBF neural network may be configured to determine, based on the following equation, a breakdown of the semiconductor equipment: Sj=exp(-∑ i=1n⁢wi*Zi,j-Xi22⁢σi2∑ i=1n⁢wi) where Sj is a modified RBF value calculated based on the measurement data, wi selects an ith component of the measurement data, σi standardizes the ith component of the measurement data, Zi,j is an ith component of standard data indicating a normal state of the semiconductor equipment, and Xi is the ith component of the measurement data. wi may have a value of 0 or 1. wi may have a value determined based on a correlation coefficient between components of the standard data. According to one or more embodiments, a processor includes a modified radial basis function (RBF) neural network configured to determine, based on measurement data with respect to semiconductor equipment, a breakdown of the semiconductor equipment. The modified RBF neural network is configured to determine, based on Equation 1 below, the breakdown of the semiconductor equipment: Sj=exp(-∑ i=1n⁢wi*Zi,j-Xi22⁢σi2∑ i=1n⁢wi)[Equation⁢ 1] where Sj is a modified RBF value calculated based on the measurement data, wi selects an ith component of the measurement data, σi standardizes the ith component of the measurement data, Zi,j is an ith component of standard data indicating a normal state of the semiconductor equipment, and Xi is the ith component of the measurement data. wi is determined by Matrix C calculated based on the standard data and Equation 2 below: C=|corr(Z,Z)| [Equation 2] wherein Matrix C is a correlation coefficient between components of the standard data. When an ith component of any one selected from among rows of Matrix C is less than or equal to a threshold value, wi may have a value of 0. When an ith component of any one selected from among rows of Matrix C is greater than or equal to a threshold value, wi may have a value of 1. The threshold value may be in a range of about 0.5 to about 0.7. σi may be a standard deviation of the i