US-12628618-B2 - Apparatus and method for setting semiconductor parameter

US12628618B2US 12628618 B2US12628618 B2US 12628618B2US-12628618-B2

Abstract

Disclosed are a method and apparatus for setting a semiconductor parameter. The method for setting a semiconductor parameter according to an embodiment of the present disclosure is a method performed on a computing apparatus including one or more processors and a memory storing one or more programs executed by the one or more processors, the method including acquiring electrical measurement parameters corresponding to preset semiconductor manufacturing parameters, classifying the electrical measurement parameters into a plurality of groups according to a degree of correlation, extracting a correlation axis reflecting a correlation between electrical measurement parameters belonging to a corresponding group for each classified group, and predicting a figure of merit of a semiconductor device by using data values of electrical measurement parameters belonging to the corresponding group as input based on the correlation axis of each group.

Inventors

Rock Hyun Baek
Hyeok YUN

Assignees

POSTECH Research and Business Development Foundation

Dates

Publication Date: 20260512
Application Date: 20221108
Priority Date: 20220225

Claims (18)

1 . A method for setting a semiconductor parameter, the method being performed on a computing apparatus including one or more processors and a memory storing one or more programs executed by the one or more processors, the method comprising: acquiring electrical measurement parameters corresponding to preset semiconductor manufacturing parameters; classifying the electrical measurement parameters into a plurality of groups according to a degree of correlation; extracting a correlation axis reflecting a correlation between electrical measurement parameters belonging to a corresponding group for each classified group; and predicting a figure of merit of a semiconductor device by using data values of electrical measurement parameters belonging to the corresponding group as input based on the correlation axis of each group, wherein the classifying of the electrical measurement parameters into the plurality of groups includes calculating variance inflation factors between the electrical measurement parameters, respectively, and classifying electrical measurement parameters of which the calculated variance inflation factor is equal to or greater than a preset threshold value into the same group.
2 . The method of claim 1 , wherein the extracting of the correlation axis includes generating an artificial neural network model for each group when grouping of the electrical measurement parameters is completed, and extracting a correlation axis between the electrical measurement parameters belonging to each group by training each of the generated artificial neural network models.
3 . The method of claim 2 , wherein the number of the correlation axes to be extracted is set to be the same as the number of electrical measurement parameters belonging to the corresponding group.
4 . The method of claim 2 , wherein the calculating of the variance inflation factor when a missing value is included in the acquired electrical measurement parameter includes checking whether or not there are data that are not missing at the same time in two electrical measurement parameters, and calculating a variance inflation factor of the two electrical measurement parameters by using the data that are not missing at the same time when there are the data that are not missing at the same time.
5 . The method of claim 2 , wherein when missing data is included in the acquired electrical measurement parameter, the method further includes training the artificial neural network model by setting a partial loss error to zero for the missing data when the artificial neural network model is trained, and predicting a missing value of a corresponding electrical measurement parameter by inputting the extracted correlation axis into the artificial neural network model when training of the artificial neural network model is completed.
6 . The method of claim 1 , further comprising: setting an effective correlation axis among the correlation axes extracted for each group, wherein in the predicting of the figure of merit of the semiconductor device, the figure of merit of the semiconductor device is predicted by using data values of electrical measurement parameters belonging to the corresponding group as input based on the effective correlation axis set in each group.
7 . The method of claim 6 , wherein the setting of the effective correlation axis includes calculating explained variances (EVs) for the correlation axes extracted for each group, respectively, and setting, among the correlation axes, a correlation axis in which a value of the calculated explained variance is equal to or greater than a preset threshold value as the effective correlation axis.
8 . The method of claim 1 , further comprising: limiting a data range of the electrical measurement parameters in each classified group within a data distribution range with the extracted correlation axis as a reference.
9 . The method of claim 8 , further comprising: calculating sensitivity to the figure of merit of the electrical measurement parameters within the data distribution range with the correlation axis as a reference; and selecting an electrical measurement parameter capable of optimizing the corresponding figure of merit based on the calculated sensitivity.
10 . An apparatus for setting a semiconductor parameter, comprising: a preprocessing module configured to acquire electrical measurement parameters corresponding to preset semiconductor manufacturing parameters, classify the electrical measurement parameters into a plurality of groups according to a degree of correlation, and extract a correlation axis reflecting a correlation between electrical measurement parameters belonging to a corresponding group for each classified group; and a prediction module configured to predict a figure of merit of a semiconductor device by using data values of electrical measurement parameters belonging to the corresponding group as input based on the correlation axis of each group, wherein the preprocessing module is configured to calculate variance inflation factors between the electrical measurement parameters, respectively, and classify electrical measurement parameters of which the calculated variance inflation factor is equal to or greater than a preset threshold value into the same group.
11 . The apparatus of claim 10 , wherein the preprocessing module is configured to generate an artificial neural network model for each group when grouping of the electrical measurement parameters is completed, and extract a correlation axis between the electrical measurement parameters belonging to each group by training each of the generated artificial neural network models.
12 . The apparatus of claim 11 , wherein the number of the correlation axes to be extracted is set to be the same as the number of electrical measurement parameters belonging to the corresponding group.
13 . The apparatus of claim 11 , wherein the preprocessing module is configured to check whether or not there are data that are not missing at the same time in two electrical measurement parameters when a missing value is included in the acquired electrical measurement parameter, and calculate a variance inflation factor of the two electrical measurement parameters by using the data that are not missing at the same time when there are the data that are not missing at the same time.
14 . The apparatus of claim 11 , wherein the preprocessing module is configured to train the artificial neural network model by setting a partial loss error to zero for missing data when the artificial neural network model is trained, and predict a missing value of a corresponding electrical measurement parameter by inputting the extracted correlation axis into the artificial neural network model when training of the artificial neural network model is completed.
15 . The apparatus of claim 10 , wherein the preprocessing module is configured to set an effective correlation axis among the correlation axes extracted for each group, and the prediction module is configured to predict the figure of merit of the semiconductor device by using data values of electrical measurement parameters belonging to the corresponding group as input based on the effective correlation axis set in each group.
16 . The apparatus of claim 15 , wherein the preprocessing module is configured to calculate explained variances (EVs) for the correlation axes extracted for each group, respectively, and set, among the correlation axes, a correlation axis in which a value of the calculated explained variance is equal to or greater than a preset threshold value as the effective correlation axis.
17 . The apparatus of claim 10 , wherein the prediction module is configured to limit a data range of the electrical measurement parameters in each classified group within a data distribution range with the extracted correlation axis as a reference.
18 . The apparatus of claim 17 , further comprising: an analysis module configured to calculate sensitivity to the figure of merit of the electrical measurement parameters within the data distribution range with the correlation axis as a reference and select an electrical measurement parameter capable of optimizing the corresponding figure of merit based on the calculated sensitivity.

Description

CROSS-REFERENCE TO RELATED APPLICATION This application claims the benefit under 35 USC § 119(a) of Korean Patent Application No. 10-2022-0025392, filed on Feb. 25, 2022, in the Korean Intellectual Property Office, the entire disclosure of which is incorporated herein by reference for all purposes. BACKGROUND 1. Field Embodiments of the present disclosure relate to a technique for setting a semiconductor parameter. 2. Description of Related Art In a semiconductor process, it is necessary to develop a semiconductor device with low cost and time and increase the mass production yield due to competition in miniaturization and integration technologies. Techniques using machine learning through an artificial neural network have been proposed as a method for reducing cost and time in a process of developing the semiconductor device. However, in an actual semiconductor process environment, the measurement of parameters is performed through sampling to reduce the process cost, and thus a missing value occurs. As a result, an inappropriate and incomplete data set with which the artificial neural network is trained is acquired. In addition, if a process recipe is set and the process is performed based on a specific target value of a semiconductor manufacturing parameter, data generated as a result of the process inevitably includes process variability and random variability that are difficult to control based on the target value. As a result, an input value with which the artificial neural network is trained is one value that is a target value of the semiconductor manufacturing parameter set by the process recipe, but an output value becomes a large number of distributed data including process variability and random variability in the target value. Accordingly, there is a problem that one-to-one data pairs used for general machine learning cannot be acquired. Here, an electrical measurement parameter (EPM) reflecting information of the process step may be acquired and used as an input value, but the electrical measurement parameters (EPMs) may have a statistical correlation with each other (e.g., in the case of resistance and capacitance, the thicker the metal, the lower the resistance but the larger the capacitance), and the correlation between the electrical measurement parameters (EPMs) is a factor that makes it impossible to perform sensitivity analysis and parameter tuning of the electrical measurement parameter (EPM) alone or independently. In addition, it is very inefficient to perform, by an engineer, the sensitivity analysis and parameter tuning while simultaneously considering the correlation of these electrical measurement parameters in the process of developing the semiconductor device because complex computation is required. In addition, in order to obtain device characteristic data according to the semiconductor manufacturing parameter, it is necessary to proceed with the semiconductor process while changing the process recipe and obtain the device characteristic data after the process is performed. Accordingly, there is a problem in that a lot of time and money is consumed to proceed with the semiconductor process while variously changing the process recipe. PRIOR ART LITERATURE Patent Literature PTL 1: Korean Unexamined Patent Publication No. 10-2019-0003909 (Jan. 10, 2019) SUMMARY Embodiments of the present disclosure are to provide a new technique for setting a semiconductor parameter. A method for setting a semiconductor parameter according to an embodiment of the present disclosure is a method performed on a computing apparatus including one or more processors and a memory storing one or more programs executed by the one or more processors, the method including acquiring electrical measurement parameters corresponding to preset semiconductor manufacturing parameters, classifying the electrical measurement parameters into a plurality of groups according to a degree of correlation, extracting a correlation axis reflecting a correlation between electrical measurement parameters belonging to a corresponding group for each classified group, and predicting a figure of merit of a semiconductor device by using data values of electrical measurement parameters belonging to the corresponding group as input based on the correlation axis of each group. The classifying of the electrical measurement parameters into the plurality of groups may include calculating variance inflation factors between the electrical measurement parameters, respectively, and classifying electrical measurement parameters of which the calculated variance inflation factor is equal to or greater than a preset threshold value into the same group. The extracting of the correlation axis may include generating an artificial neural network model for each group when grouping of the electrical measurement parameters is completed, and extracting a correlation axis between the electrical measurement parameters belonging to each group by training e