CN-116127835-B - Self-adaptive optimization method for optical scattering measurement

Abstract

The invention discloses an adaptive optimization method of optical scattering measurement, which comprises the steps of carrying out parameterization modeling on a nano structure to be measured, building a basic coarse-granularity sample library, training and building a coarse-granularity proxy model based on the sample library, completing rapid identification of a multi-point area through the proxy model, carrying out adaptive sampling and library building according to a key area, wherein the generated sample library is a fine-granularity sample library, training and building a fine-granularity proxy model based on the sample library, completing fine-granularity search optimization in the key area through the model, and searching for optimal values in the multi-point area, wherein the optimal values are relevant parameters of the sample to be measured. According to the invention, the searching range can be reduced through the variable-granularity proxy model, the library can be built quickly, the optimizing efficiency is improved through the self-adaptive optimizing strategy, and further, the related parameters of the sample to be measured can be solved accurately and efficiently.

Inventors

• MENG KAI
• XU YIHANG
• LOU PEIHUANG
• QIAN XIAOMING
• WU XING

Assignees

• 南京航空航天大学
• 南京航空航天大学苏州研究院

Dates

Publication Date

20260505

Application Date

20221220

Claims (1)

1. 1. An adaptive optimization method for optical scatterometry, comprising the steps of: Step 1, carrying out parameterized modeling on a specific semiconductor nano structure according to process requirements to generate a coarse-granularity optical scattering response sample library alpha; the input parameters of the coarse-granularity optical scattering response sample library comprise a nanostructure, material parameters, process nodes and measurement configuration parameters, wherein the expression form of the output scattering response of the coarse-granularity optical scattering response sample library comprises the reflectivity, diffraction efficiency, jones matrix and Mueller matrix of the structure to be detected; The nano structure comprises a line width, a line height and a side wall angle, wherein the material parameters comprise a refractive index n and an extinction coefficient k; Step 2, constructing a coarse-granularity proxy model S1 based on optical scattering response sample library alpha training; taking an optical scattering response sample library alpha as a sampling space for sampling design, then generating a training set, and constructing a coarse-grained proxy model S1 by training the training set; the sampling method adopted in the step comprises Latin hypercube sampling LHD or uniform design UD; Step 3, global coarse search is carried out based on the coarse-grained proxy model S1, and rapid identification of a plurality of key areas is completed; Step 4, performing fine granularity rapid library establishment based on the parameter variation range to be detected in the key area to generate an optical scattering response sample library beta; Step 5, constructing a fine granularity agent model S2 based on optical scattering response sample library beta training; Taking an optical scattering response sample library beta as a sampling space for sampling, then generating a training set, constructing a fine-granularity agent model S2 by training, and adopting a self-adaptive sampling method based on variance, cross validation error and local gradient; and 6, performing fine search on the key areas based on the agent model S2, wherein the optimal values in the key areas are finally output parameters to be measured.

Description

Self-adaptive optimization method for optical scattering measurement Technical Field The invention belongs to the technical field of computational optical measurement, and relates to a method for realizing rapid inverse determination of parameters to be measured and rapid configuration optimization of optical scattering measurement, which can be applied to rapid measurement of geometric quantity and physical quantity of a semiconductor nano structure based on optical scattering. Background Semiconductor and integrated circuit fabrication necessarily requires quantity inspection during the fabrication process flow. Currently, semiconductor front-end metrology techniques mainly include scanning electron microscopy (Scanning Electron Microscope, SEM), atomic force microscopy (Atomic Force Microscope, AFM), transmission electron microscopy (Transmission Electron Microscope, TEM), and optical scatterometry (Optical Scatterometry) techniques. The optical scattering measurement refers to projecting a beam of polarized light onto the surface of a sample to be measured, and solving the relevant dimensional parameters of the sample to be measured by measuring the changes of the polarization states before and after reflection. Because the optical scattering measurement has the advantages of rapidness, no damage and the like, the method is a mainstream technical scheme of the current online optical measurement (especially in the field of optical critical dimension measurement). The spectroscopic ellipsometry is not limited by the optical diffraction limit, is a spectroscopic, computational and model-based indirect measurement method, and is widely applied to the aspects of predictive approximation, optimization design and the like in the technical field of computational optical measurement. Currently, optical scatterometry mainly includes two key techniques, forward modeling and inverse solution. Forward modeling of the optical response of the nanostructure is mainly achieved through methods such as strict coupled wave analysis (Rigorous coupled WAVE ANALYSIS, RCWA), finite element analysis (FINITE ELEMENT ANALYSIS, FEA) and time domain finite difference method (FINITE DIFFERENCE TIME domain, FDTD). And the inverse solution mainly comprises two methods, namely library search matching and direct fitting optimization technology. The library search matching method is simple to operate and easy to implement, but a scattering measurement ellipsometry response simulation sample library needs to be established in advance according to the change range of input parameters, which requires a great deal of offline modeling and simulation operation work. And the performance of the library search matching method depends greatly on the network granularity of the ellipsometric simulation sample library and the relative interpolation algorithm thereof. The direct fitting optimization technology does not need to establish a simulation sample library in advance, but each iteration needs to call forward modeling to calculate, so that repeated calling and solving are needed, further, the simulation calculation efficiency is low, and the calculation cost is high. Therefore, the optical scattering measurement simulation calculation performance and the simulation calculation efficiency of the semiconductor nano structure at the present stage are one of main reasons for limiting the fast fitting optimization technology. A library matching method based on fitting error interpolation for optical scatterometry is proposed in CN 102798342B. The method utilizes a spectrum library established in advance and a spectrum response obtained by measurement to construct an interpolation function of fitting errors, and further converts an inverse solving problem of a sample structure to be measured in optical scattering measurement into an optimal value problem for solving the interpolation objective function of the fitting errors. The optimal value of the objective function is the geometric parameter of the sample to be measured. But this method still requires the creation of a complete library of spectral responses prior to measurement, which requires extensive prior simulation work. Meanwhile, the final measurement accuracy of the method not only depends on the granularity of the established spectral response library, but also depends on the performance of the interpolation algorithm used. CN103559329B proposes an extraction method for characteristic parameters of coarse nanostructures in optical scatterometry. The method determines the mapping relation between the extracted parameters of the coarse nano structure and the parameters to be measured through simulation calculation, and optimizes the measurement configuration and the equivalent medium model. And then, on the basis of the simulation result, correcting the parameter to be detected by using a parameter mapping relation, thereby obtaining the characteristic parameter of the coarse nano structure