Tokyo Electron Limited

Kla Tencor Corporation

Timbre Technologies, Inc.

General Electric Company

Ge Precision Healthcare LLC

Wen Jin

Junwei Bao

Shifang Li

Vi Vuong

Gopal Biligeri Avinash

Leonid Poslavsky

Manuel B Madriaga

Srinivas Rao Doddi

Min Zhang

Lie-Quan Lee

Zili Ma

Ping Wang

Walter Dean Mieher

Xinhui Niu

John Josef Hench

Hanyou Chu

Xinkang Tian

Joerg Bischoff

Emmanuel Drege

Ching-Ling Meng

Holger Tuitje

Mihail Mihaylov

Weiwen Xu

Karl Hehl

Kevin H Leung

Techniques are provided for determining confident data samples for machine learning (ML) models on unseen data. In one embodiment, a method is provided that comprises extracting, by a system comprising a processor, a feature vector for a data sample based on projection of the data sample onto a standard feature space. The method further comprises processing, by the system, the feature vector using an outlier detection model to determine whether the data sample is within a scope of a training dataset used to train a machine learning model, wherein the outlier detection model was trained using features extracted from the training dataset based on projection of data samples included in the training dataset onto the standard feature space.

Determining confident data samples for machine learning models on unseen data

Techniques are provided for evaluating and defining the scope of data-driven deep learning models. In one embodiment, a machine-readable storage medium is provided comprising executable instructions that, when executed by a processor, facilitate performance of operations comprising employing a machine learning model to extract first training data features included in a training data set and first target data features included in a target data set. The operations further comprise determining whether the target data set is within a defined data scope of the training data set based on analysis of correspondences between the first training data features and the first target data feature, and determining whether application of the target data set to a target neural network model developed using the training data set will generate results with an acceptable level of accuracy based on whether the target data set is within the defined data scope.

Data-driven deep learning model generalization analysis and improvement

Embodiments are generally directed to neural network training for library-based critical dimension metrology. An embodiment of a method includes optimizing a threshold for a principal component analysis of a spectrum data set to provide a principal component value, estimating a training target for one or more neural networks, training the one or more neural networks based both on the training target and on the principal component value provided from optimizing the threshold for the principal component analysis, and providing a spectral library based on the one or more trained neural networks.

Accurate and fast neural network training for library-based critical dimension (CD) metrology

Described is a method and system for measuring parameters of a structure on a substrate, such as a via or a through-silicon via (TSV) using an imaging X-ray metrology system. A previously-trained Support Vector Machine (SVM) model is used to extract structure parameters from the acquired structure X-ray images. Training of the Support Vector Machine (SVM) model is accomplished by using a library of actual or simulated X-ray images, or a combination of the two image types, paired with structure parameter sets.

Nondestructive inline X-ray metrology with model-based library method

Disclosed is an in-situ optical monitor (ISOM) system and associated method for controlling plasma etching processes during the forming of stepped structures in semiconductor manufacturing. The in-situ optical monitor (ISOM) can be optionally configured for coupling to a surface-wave plasma source (SWP), for example a radial line slotted antenna (RLSA) plasma source. A method is described to correlate the lateral recess of the steps and the etched thickness of a photoresist layer for use with the in-situ optical monitor (ISOM) during control of plasma etching processes in the forming of stepped structures.

System for in-situ film stack measurement during etching and etch control method

Approaches for accurate neural network training for library-based critical dimension (CD) metrology are described. Approaches for fast neural network training for library-based CD metrology are also described. In an example, a method includes optimizing a threshold for a principal component analysis (PCA) of a spectrum data set to provide a principal component (PC) value, estimating a training target for one or more neural networks, training the one or more neural networks based both on the training target and on the PC value provided from optimizing the threshold for the PCA, and providing a spectral library based on the one or more trained neural networks.

Automated determination of a number of profiles for a training data set to be used in training a machine learning system for generating target function information from modeled profile parameters. In one embodiment, a first principal component analysis (PCA) is performed on a training data set, and a second PCA is performed on a combined data set which includes the training data set and a test data set. A test data set estimate is generated based on the first PCA transform and the second PCA matrix. The size of error between the test data set and the test data set estimate is used to determine whether a number of profiles associated with the training data set is sufficiently large for training a machine learning system to generate a library of spectral information.

Determination of training set size for a machine learning system

Methods of generating wide process range libraries for metrology are described. For example, a method includes generating a first library having a first process range for a first parameter. A second library is generated having a second process range for the first parameter. The second process range is overlapping with the first process range. The second library is stitched to the first library to generate a third library having a third process range for the first parameter. The third process range is wider than each of the first and second process ranges.

Wide process range library for metrology

Eigensolutions for use in determining the profile of a structure formed on a semiconductor wafer can be approximated by obtaining a known set of eigenvectors associated with a first section of a hypothetical profile of the structure, where the known set of eigenvectors is used to generate a simulated diffraction signal for the hypothetical profile. A known characteristic matrix associated with a second section of a hypothetical profile is obtained, and an approximated set of eigenvalues for the second section is determined based on the known set of eigenvectors associated with the first section and the known characteristic matrix associated with the second section.

Approximating eigensolutions for use in determining the profile of a structure formed on a semiconductor wafer

A fabrication tool can be controlled using a support vector machine. A profile model of the structure is obtained. The profile model is defined by profile parameters that characterize the geometric shape of the structure. A set of values for the profile parameters is obtained. A set of simulated diffraction signals is generated using the set of values for the profile parameters, each simulated diffraction signal characterizing the behavior of light diffracted from the structure. The support vector machine is trained using the set of simulated diffraction signals as inputs to the support vector machine and the set of values for the profile parameters as expected outputs of the support vector machine. After the support vector machine has been trained, a fabrication process is performed using the fabrication tool to fabricate the structure on the wafer. A measured diffraction signal off the structure is obtained. The measured diffraction signal is inputted into the trained support vector machine. Values of profile parameters of the structure are obtained as an output from the trained support vector machine. One or more process parameters or equipment settings of the fabrication tool are adjusted based on the obtained values of the profile parameters.

Controlling a fabrication tool using support vector machine

A structure formed on a semiconductor wafer can be examined using a support vector machine. A profile model of the structure is obtained. The profile model is defined by profile parameters that characterize the geometric shape of the structure. A set of values for the profile parameters is obtained. A set of simulated diffraction signals is generated using the set of values for the profile parameters, each simulated diffraction signal characterizing the behavior of light diffracted from the structure. The support vector machine is trained using the set of simulated diffraction signals as inputs to the support vector machine and the set of values for the profile parameters as expected outputs of the support vector machine. A measured diffraction signal off the structure is obtained. The measured diffraction signal is inputted into the trained support vector machine. Values of profile parameters of the structure are obtained as an output from the trained support vector machine.

Optical metrology using a support vector machine with simulated diffraction signal inputs

A structure formed on a semiconductor wafer can be examined using a support vector machine. A profile model is defined by profile parameters that characterize the geometric shape of the structure. A training set of values for the profile parameters is obtained. A training set of simulated diffraction signals is generated using the training set of values for the profile parameters. The support vector machine is trained using the training set of values for the profile parameters. A simulated diffraction signal is generated using a set of values for the profile parameters as inputs to the trained support vector machine. A measured diffraction signal is compared to the simulated diffraction signal. When the signals match within one or more matching criteria, values of profile parameters of the structure are determined to be the set of values for the profile parameters used to generate the simulated diffraction signal.

Optical metrology using support vector machine with profile parameter inputs

The number of diffraction orders to use in generating simulated diffraction signals for a two-dimensional structure in optical metrology is selected by generating a first simulated diffraction signal using a first number of diffraction orders and a hypothetical profile of the two-dimensional structure. A second simulated diffraction signal is generated using a second number of diffraction orders using the same hypothetical profile used to generate the first simulated diffraction signal, where the first and second numbers of diffraction orders are different. The first and second simulated diffraction signals are compared. Based on the comparison of the first and second simulated diffraction signals, a determination is made as to whether to select the first or second number of diffraction orders.

Optimization of diffraction order selection for two-dimensional structures

A profile parameter value is determined in integrated circuit metrology by: a) determining a diffraction signal difference based on a measured diffraction signal and a previously generated diffraction signal; b) determining a first profile parameter value based on the previously generated diffraction signal; c) determining a first profile parameter value change based on the diffraction signal difference; d) determining a second profile parameter value based on the first profile parameter value change; e) determining a second profile parameter value change between the first and second profile parameter values; f) determining if the second profile parameter value change meets one or more preset criteria; and g) when the second profile parameter value change fails to meet the one or more preset criteria, iterating c) to g) using as the diffraction signal difference in the iteration of step c), a diffraction signal difference determined based on the measured diffraction signal and a diffraction signal for the second profile parameter value previously determined in step d), and as the first profile parameter value in the iteration of step e), the second profile parameter value previously determined in step d).

Growing community of inventors

Fremont, CA, United States of America

Wen Jin