Method for identifying semiconductor nano structure pattern

Abstract

The invention discloses a method for identifying a nano structure pattern based on support vector machines, which comprises the following steps: firstly generating a training spectrum, and determining a kernel function and a training mode of the support vector machine; generating a test spectrum and multiple different support vector machines; carrying out feature pattern identification accuracy testing on the support vector machines by using the test spectrum, and finding out the relationship among the identification accuracy, the number of training spectra and the kernel function as the guiding principle of training of the support vector machines; adding different magnitudes of noise effects to the test spectrum, applying the test spectrum containing different magnitude of noises into the support vector machines to carry out testing, and finding out the maximum noise magnitude, which can be added on the premise of ensuring high correct recognition rate, as the other guiding principle; training by using the two guiding principles to obtain the optimal support vector machine; and mapping the measurement spectrum corresponding to the true structure to be identified, thereby identifying the pattern. The invention can be used for quickly and accurately identifying the pattern feature of the semiconductor nano structure.

Description

technical field [0001] The invention belongs to the field of semiconductor scattering optics measurement, and in particular relates to a nanostructure shape recognition method based on a support vector machine, which is suitable for quickly and accurately recognizing the shape characteristics of semiconductor nanostructures. Background technique [0002] In the field of measuring the characteristic dimensions of semiconductor nanostructures based on optical principles, optical scatterometer (Scatterometry) is the most common characteristic dimension measurement equipment. The measurement process includes two parts: forward modeling of optical properties and reverse calculation. The forward optical characteristic modeling is to carry out the optical scattering field simulation on the geometric model of the nanostructure to be tested through the forward optical characteristic modeling program, and obtain the simulated spectrum. The reverse calculation part includes continuous...

AI Technical Summary

Problems solved by technology

However, the neural network is designed based on the principle of empirical risk minimization. This innate lack of involvement leads to the lack of its generalization ability, that is, when the real shape parameter range of the structure to be tested exceeds the structural parameters corresponding to the training spectrum When the value range is set, the classification recognition accuracy rate will not be guaranteed

Images