Search

CN-121994149-A - Method for predicting surface shape and thickness change of multilayer element by multi-surface interference chromatography learning

CN121994149ACN 121994149 ACN121994149 ACN 121994149ACN-121994149-A

Abstract

The invention relates to the technical field of image processing, in particular to a method for predicting the surface shape and thickness change of a multilayer element by multi-surface interference chromatography learning, which comprises the steps of constructing the surface shape distribution of each surface shape of a two-layer element by utilizing Zernike polynomials; the method comprises the steps of constructing a multi-surface interferogram based on an interference theory model, training a chromatographic neural network model by utilizing an initial multi-surface interferogram based on two layers of elements and a corresponding phase shift interferogram to separate a single-surface interferogram, training a two-step phase shift unsupervised learning model, further processing the two-frame phase shift single-surface interferogram, and outputting a wrapping phase. The invention has noise resistance, and can realize high-precision multilayer optical element even extend to multilayer film global measurement by a small amount of interference patterns.

Inventors

  • Yan Ketao
  • LIU HONGYI
  • HUANG XINHAO
  • ZHOU YONGHAO
  • YU YINGJIE

Assignees

  • 常州大学

Dates

Publication Date
20260508
Application Date
20260325
Priority Date
20251014

Claims (10)

  1. 1. The method for predicting the surface shape and thickness change of the multilayer element by multi-surface interference chromatography learning is characterized by comprising the following steps: Constructing each surface profile distribution of the two-layer element by utilizing Zernike polynomials; training the chromatographic neural network model by utilizing the interference patterns of the two layers of elements and the corresponding phase shift interference patterns, and separating out single-surface element interference signals; training a two-step phase shift unsupervised learning model, further processing the two-frame phase shift single-surface interferogram, and outputting a wrapping phase and a phase shift estimated value.
  2. 2. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning according to claim 1, wherein the model formula of the interferogram is: (3) Wherein I 1 (x, y) represents an initial multi-surface interference pattern, A (x, y) and B (x, y) are background light intensity and modulation degree, m represents that 6 interference signals are overlapped in multi-surface interference signals of two layers of elements, and lambda 0 is an initial wavelength.
  3. 3. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning of claim 1, wherein the chromatographic neural network model comprises: The method comprises the steps of inputting a feature map into a convolution extraction layer, mapping the number of channels to 64, sequentially increasing the number of channels to 128, 256, 512, 1024 and 2048 after five times of downsampling by each downsampling layer, entering up-sampling after once convolving the extraction layer by a bottleneck layer, splicing and fusing the up-sampled output and the corresponding downsampled feature map through jump connection, sequentially reducing the number of channels to 1024, 512, 256, 128 and 64 after five times of upsampling by one convolution extraction layer, and mapping the number of channels to 12 channel feature maps through a 1X 1 convolution output layer.
  4. 4. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning of claim 3, wherein the convolution extraction layer comprises performing residual error on an input feature map after the input feature map sequentially passes through two stacked CBL modules (Conv-BN-LeakyReLU).
  5. 5. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning according to claim 1, wherein the formula of the mixed loss function of the chromatographic neural network model is: Wherein, the 、 The MSE is the mean square error loss, GL is the gradient loss, for the weights.
  6. 6. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning according to claim 1, wherein the background light intensity a is extracted from an initial interferogram (X, y) and modulation degree B (X, y) reconstructing interferogram I using wrapped phase and phase shift estimates 1 (x, y)、I 2 (X, y) calculating a single surface initial interferogram I 1 ' (x, y) and I 1 Loss of (x, y) Loss 1 , phase shift interferograms I 2 ' (x, y) and I are calculated 2 Loss of (x, y) Loss 2 the two-step phase shift unsupervised learning model weights were optimized using total loss=loss 1 +Loss 2 back propagation.
  7. 7. The method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning according to claim 1, wherein the two-step phase shift unsupervised learning model is evaluated using an RMS index of root mean square error.
  8. 8. The method of claim 1 wherein the interferogram is a 6-set two-frame initial single-surface interferogram comprising a reference surface interferogram with a first element front surface interferogram, a reference surface interferogram with a contact surface interferogram with a second element back surface interferogram, a first element front surface interferogram with a contact surface interferogram, a second element front surface interferogram with a second element back surface interferogram, and a two element contact surface interferogram with a second element back surface interferogram.
  9. 9. A system for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning, comprising a memory for storing instructions executable by a processor, and a processor for executing the instructions to implement the method for predicting surface shape and thickness variation of a multilayer element by multi-surface interferometry learning of any one of claims 1-8.
  10. 10. A computer readable medium storing computer program code, wherein the computer program code when executed by a processor implements a method of multi-surface interferometry learning to predict surface shape and thickness variations of a multilayer element according to any of claims 1-8.

Description

Method for predicting surface shape and thickness change of multilayer element by multi-surface interference chromatography learning Technical Field The invention relates to the technical field of image processing, in particular to a method for predicting the surface shape and thickness change of a multilayer element by multi-surface interference chromatography learning, which can be used for measuring the surface shape and thickness of a multilayer structure such as a precision optical element, a film and the like. Background The multilayer element is widely applied to components such as integrated circuits, semiconductor chips, optical devices and the like, and physical parameters such as thickness, three-dimensional contour and the like of each layer of element influence the performance and reliability of the components, so that the realization of high-precision and global measurement of surface shape and thickness variation parameters in the multilayer element is very important. The optical measurement method is applied to measurement of the surface shape or thickness of an element, and according to different principles, the classical method mainly comprises an ellipsometry method, a white light interferometry method, a spectroscopic method and a spectroscopic method, wherein the ellipsometry method utilizes the polarization change of reflected light after the polarized light is reflected by the element to analyze the film thickness, the optical constants and the interface characteristics, the white light interferometry method searches the peak value of interference fringes by changing the optical path difference in the vertical direction, the surface morphology and the thickness can be reconstructed after the relative height is solved, and the spectroscopic method inverts the optical constants and the thickness changes by analyzing the spectrum reflected by or transmitted by the element. In order to further improve the detection precision, technology evolution upgrade and fusion of various methods realize effective measurement. The method has the advantages of certain precision under specific application conditions, but also has the limitations that the system stability requirement is high due to the fact that an elliptical polarization method is complex in experimental light path, the thickness and the refractive index are the main concerns, the white light interferometry needs vertical scanning, the measurement speed is low, the white light interferometry is sensitive to environmental vibration, the spectrometry is simple in operation, the measurement precision is easy to be interfered by the environment, and the like. Disclosure of Invention The invention provides a method for predicting the surface shape and thickness change of a multilayer element by multi-surface interference chromatography learning, which aims at overcoming the defects of the existing method. The method comprises the steps of designing a multi-surface interference chromatographic neural network model, inputting two frames of multi-surface phase shift interferograms, outputting separated single-surface element interference signals, adopting a label supervision fit loss function to iteratively optimize errors between the predicted interferograms and true interferograms, finally, enabling the model to realize chromatographic interference signal separation of the multi-surface interferograms, training a two-step phase shift non-supervision learning model, inputting two frames of single-surface phase shift interferograms, reconstructing the interferograms by using wrapping phase and phase shift estimated values calculated by a network, calculating the loss function with the input interferograms, reversely propagating and optimizing network weights, realizing non-supervision learning based on a physical model, and recovering wrapping phases based on element surface shapes from the multi-surface interferograms based on the two models. The technical scheme adopted by the invention is that the method for predicting the surface shape and thickness change of the multilayer element by multi-surface interference chromatography learning comprises the following steps: Constructing each surface profile distribution of the two-layer element by utilizing Zernike polynomials; as a preferred embodiment of the present invention, the model formula of the multi-surface interferogram is: (3) Wherein I 1 (x, y) represents an initial multi-surface interference pattern, A (x, y) and B (x, y) are background light intensity and modulation degree, m represents that 6 interference signals are overlapped in multi-surface interference signals of two layers of elements, and lambda 0 is an initial wavelength. As a preferred embodiment of the invention, the interferograms are 6 sets of two-frame initial single-surface interferograms, wherein the initial single-surface interferograms comprise reference surface interferograms and first element (element 1) front surf