Method for extracting critical dimension of semiconductor nanostructure

A method for extracting a critical dimension of a semiconductor nanostructure, the method comprising:

• 1) determining a value range for each parameter to be extracted, whereby generating an electronic spectra database, and employing training spectra and support vector machine (SVM) training networks for training of SVMs;
• 2) employing said SVMs after training to map measured spectra to yield a corresponding electronic spectra database;
• and
• 3) employing a searching algorithm to search for an optimum simulation spectrum in said corresponding electronic spectra database, simulation parameters corresponding to said optimum simulation spectrum being the critical dimension of the semiconductor nanostructure to be extracted;
• wherein in the process of extracting of said training spectra, each parameter to be extracted corresponds to a SVM, the number of categories contained by an output end of said SVM is determined by that of sub-ranges divided from said value range of each parameter to be extracted, each category is identified by a unique number and used for denoting a sub-range, and corresponds to a training spectra set, and each training spectra set corresponds to a sub-range, by randomly choosing multiple discrete values in a sub-range, and generating a simulation spectrum for each discrete point via said forward optical modeling program, a simulation spectra set, or a training spectra set is generated, and all training spectra contained in training spectra sets corresponding to all output ends of said SVM are training spectra required by said SVM.

A method for extracting a critical dimension of a semiconductor nanostructure, the method comprising:

• 1) determining a value range for each parameter to be extracted, whereby generating an electronic spectra database, and employing training spectra and support vector machine (SVM) training networks for training of SVMs;
• 2) employing said SVMs after training to map measured spectra to yield a corresponding electronic spectra database;
• and
• 3) employing a searching algorithm to search for an optimum simulation spectrum in said corresponding electronic spectra database, simulation parameters corresponding to said optimum simulation spectrum being the critical dimension of the semiconductor nanostructure to be extracted;
• wherein in step (2), an output end of a SVM corresponding to said parameter to be extracted comprises multiple categories, each category corresponds to a sub-range divided from a value range of said parameter to be extracted, by inputting said measured spectra into a SVM, an output of said SVM comprises a category indicating a sub-range of a value range to be extracted corresponding to said SVM, a sub-range of each parameter to be extracted is determined by mapping of a SVM corresponding to each parameter to be extracted, and all sub-ranges uniquely determine said electronic spectra database.

A method for extracting a critical dimension of a semiconductor nanostructure, the method comprising:

• 1) determining a value range for each parameter to be extracted, whereby generating an electronic spectra database, and employing training spectra and support vector machine (SVM) training networks for training of SVMs;
• 2) employing said SVMs after training to map measured spectra to yield a corresponding electronic spectra database;
• and
• 3) employing a searching algorithm to search for an optimum simulation spectrum in said corresponding electronic spectra database, simulation parameters corresponding to said optimum simulation spectrum being the critical dimension of the semiconductor nanostructure to be extracted;
• wherein said searching for an optimum simulation spectrum comprises:
• a) arranging all spectra in each electronic spectra database into a matrix, each row of said matrix being a simulation spectrum, and said simulation spectrum uniquely corresponding to a group of parameter values;
• b) calculating evaluation function values with each simulation spectrum and said measured spectra in said matrix from the top down according to an evaluation function;
• and
• c) searching for a minimum evaluation function value via said searching algorithm or a sorting algorithm, a simulation spectrum corresponding to said minimum evaluation function value being said optimum simulation spectrum.