A model construction method for nondestructive detection of optical constants and thickness of multilayer films

Abstract

The present invention discloses a method for constructing a model for nondestructive testing of the optical constants and thickness of multilayer thin films, belonging to the field of terahertz measurement technology. The method comprises: constructing a nondestructive testing model including a reverse module and a forward module; the reverse module inputs a first ellipsometric spectrum and a first supplementary spectrum, and outputs the refractive index, extinction coefficient, and thickness of the thin film, which are used as inputs to the forward module; the forward module outputs a second ellipsometric spectrum and a second supplementary spectrum; preprocessing a simulated measurement data set to obtain a degraded measurement data set; inputting the ellipsometric spectrum and supplementary spectrum in the degraded measurement data set into the reverse module, and training the nondestructive testing model with the goal of minimizing the error between the ellipsometric spectrum and supplementary spectrum output by the forward module and the ellipsometric spectrum and supplementary spectrum in the degraded measurement data set. The method fully utilizes multi-beam interference information to increase the amount of information carried by the signal from the object to be measured, thereby achieving better multilayer film measurement accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of terahertz measurement technology, and more specifically, relates to a model construction method for non-destructive detection of optical constants and thickness of multilayer thin films. Background Art

[0002] Optical constants are crucial for evaluating the material properties of optoelectronic devices, such as specialty coatings and integrated circuits. Furthermore, as a means of characterizing the macroscopic and microscopic properties of materials, optical constants provide a reliable basis for analyzing electronic structure, doping concentration, and polymer properties. Optical scattering measurement is a non-contact method for measuring the optical constants of thin films. Ellipsometry, a technique within optical scattering measurement, determines the optical constants of thin films by detecting changes in the polarization state of incident and reflected light relative to the film. Due to its unique measurement characteristics, it is now widely used in scientific research and industrial production. The forward physical process of ellipsometry involves measuring the ratio of the complex reflection coefficients of parallel-polarized p-light and perpendicularly-polarized s-light to obtain the real amplitude ratio Ψ and imaginary phase difference Δ of the ellipsometric parameters. In actual ellipsometer measurements, the reverse process is used, representing an inverse problem, which is inherently more difficult to solve than the forward problem. Furthermore, such inverse problems are nonlinear regression problems and lack analytical solutions.

[0003] Traditional inverse problem-solving methods struggle to solve these types of optical scattering measurement problems. Because they rely on trial-and-error learning based on human expertise, they require a good initial guess to achieve convergence. Consequently, the solution to the ellipsometric problem is non-unique. In addition to the Cauchy dispersion formula, other commonly used dispersion formulas, such as the Hartmann, Searlemeier, and Herzberg dispersion formulas, are used to characterize material properties. To solve specific practical problems, the extensive experience of human experts is required to optimally select one from these. Furthermore, model analysis often requires combining multiple discrete models to accurately describe the sample, resulting in a plethora of initial parameters. Traditional fitting methods struggle to converge on the correct solution, forcing the entire process to be repeated until the correct solution is found. This traditional approach makes ellipsometric measurements time-consuming and inefficient. Previous studies have shown that supplementing the data with auxiliary measurements of light intensity transmission can eliminate these mathematical ambiguities and produce a unique solution. However, the introduction of additional data also introduces new challenges. The added information exacerbates the shortcomings of traditional methods, increases the complexity of the fit, and reduces their practicality. In addition, there are no broadband wave plates in the terahertz band, which means that the terahertz ellipsometer can only use a single wavelength for measurement. If the traditional ellipsometric measurement method is used for measurement, the measurement accuracy of the terahertz ellipsometer will be greatly reduced. Summary of the Invention

[0004] In response to the defects of the existing technology and the need for improvement, the present invention provides a model construction method for non-destructive detection of optical constants and thickness of multilayer films, which aims to solve the problem of low measurement accuracy existing in traditional ellipsometry measurement methods.

[0005] To achieve the above-mentioned objectives, according to one aspect of the present invention, a method for constructing a model for non-destructive testing of optical constants and thickness of multilayer thin films is provided, comprising: constructing a non-destructive testing model; wherein the non-destructive testing model comprises an inverse module and a forward module; the input of the inverse module is a first ellipsometric spectrum and a first supplementary spectrum, and the output is the refractive index, extinction coefficient and thickness of the thin film; the output of the inverse module serves as the input of the forward module, and the output of the forward module is a second ellipsometric spectrum and a second supplementary spectrum; pre-processing a simulation measurement data set to obtain a degraded measurement data set, wherein the samples in the degraded measurement data set include an ellipsometric spectrum and a supplementary spectrum; inputting the ellipsometric spectrum and the supplementary spectrum in the degraded measurement data set into the inverse module, and training the non-destructive testing model with the goal of minimizing the error between the second ellipsometric spectrum and the second supplementary spectrum output by the forward module and the ellipsometric spectrum and the supplementary spectrum in the degraded measurement data set.

[0006] Furthermore, the objectives when training the nondestructive testing model are:

[0007] (n * ,k * ,d * )=argmin n,k,d [γ||F(n,k,d)-(ψ,Δ)||2+(1-γ)||G(n,k,d)-(R,A)||2]

[0008] Among them, n * 、k * d * are the optimal n, k, and d, respectively. n, k, and d are the refractive index, extinction coefficient, and thickness of the film, respectively. γ is the weight coefficient. ψ is the amplitude ratio of p-light and s-light. Δ is the phase difference between p-light and s-light. ψ and Δ constitute the ellipsometric spectrum. R is the reflectivity of the film. A is the absorptivity of the film. R and A constitute the complementary spectrum. F() is the first function, G() is the second function, and ||||2 is the Euclidean distance.

[0009] Furthermore, F() is:

[0010]

[0011] G() is:

[0012]

[0013] A=1-R

[0014] Among them, r p is the p light reflection coefficient, r s is the light reflection coefficient.

[0015] Furthermore, the reverse module and the forward module both include several cascaded U-net units; the U-net unit includes a first convolutional layer, a second convolutional layer, a third convolutional layer and an upsampling layer; the first convolutional layer includes two residually connected convolutional units, and the second convolutional layer includes two residually connected convolutional units; a convolutional unit of the first convolutional layer, a convolutional unit of the second convolutional layer, the third convolutional layer, the upsampling layer, another convolutional unit of the second convolutional layer and another convolutional unit of the first convolutional layer are connected in sequence.

[0016] Furthermore, in the reverse module and the forward module, the number of U-net units is 3.

[0017] Furthermore, the preprocessing of the simulation measurement data set specifically includes: performing scaling processing, vector offset processing and noise addition processing on the simulation measurement data set in sequence.

[0018] According to another aspect of the present invention, a method for non-destructive testing of the optical constants and thickness of multilayer films is provided, wherein the ellipsometric spectrum and supplementary spectrum of the film to be tested are input into the trained non-destructive testing model to obtain the refractive index, extinction coefficient and thickness of the film to be tested.

[0019] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; a memory storing a computer-executable program, wherein when the program is executed by the processor, the processor executes the model building method for non-destructive detection of optical constants and thickness of multilayer thin films as described above, or executes the method for non-destructive detection of optical constants and thickness of multilayer thin films as described above.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, it implements the model construction method for non-destructive detection of optical constants and thickness of multilayer thin films as described above, or implements the method for non-destructive detection of optical constants and thickness of multilayer thin films as described above.

[0021] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0022] (1) A model construction method for non-destructive testing of optical constants and thickness of multilayer thin films is provided. Ellipsometry measurement is performed based on deep learning, the physical consistency of optical scattering is utilized, and ellipsometric measurement physical modeling is performed based on neural networks to solve the terahertz ellipsometric measurement problem. Specifically, a non-destructive testing model consisting of a cascade of inverse modules and forward modules is designed. In terms of parameter selection, the reflection absorption spectrum based on light intensity (i.e., supplementary spectrum) is supplemented on the basis of the ellipsometric parameters to enrich the data and reduce the ambiguity of the inversion problem. The spectrum is analyzed to obtain a unique solution for the optical constants and thickness of the measured film, thereby better improving the measurement accuracy.

[0023] (2) An optimal architecture of a forward module and a reverse module is provided. The SRUM structure in the forward module and the reverse module is the same. When executing different module tasks, only the number of input and output characteristic data elements is different. The architecture is simple and can achieve good forward solution accuracy and reverse solution accuracy, thereby improving measurement accuracy.

[0024] (3) This method can achieve instantaneous and accurate calculation of the results without the intervention of human experts' prior knowledge, has good analytical performance, and breaks through the limitations of the terahertz band that require multi-wavelength measurement and no broadband response polarization wave plates, paving the way for automatic, efficient and high-throughput optical characterization of optical films. This is of great significance for high-precision real-time non-destructive measurement of thin films, and is of great significance for the development of terahertz ellipsometers towards intelligent, fully automatic non-destructive measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flow chart of a method for constructing a model for nondestructive testing of optical constants and thickness of multilayer thin films provided in an embodiment of the present invention;

[0026] Figure 2 The forward physical process of ellipsometry provided by the embodiment of the present invention;

[0027] Figure 3 A flow chart of a nondestructive testing model provided by an embodiment of the present invention;

[0028] Figure 4 The stacked residual U module (SRUM) provided by an embodiment of the present invention;

[0029] Figure 5 A diagram of an experimental setup for nondestructive testing of optical constants and thickness of multilayer thin films provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0031] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0032] Example 1

[0033] A model construction method for non-destructive detection of optical constants and thickness of multilayer films, Figure 1 The process of this method is shown in Figure 1 , combined with Figure 2-Figure 5 , a model construction method for non-destructive detection of optical constants and thickness of multilayer thin films in this embodiment is described in detail, and the method includes operations S1 to S3.

[0034] Operation S1, constructing a nondestructive testing model; wherein the nondestructive testing model includes an inverse module and a forward module; the input of the inverse module is the first ellipsometric spectrum and the first supplementary spectrum, and the output is the refractive index, extinction coefficient and thickness of the film; the output of the inverse module is used as the input of the forward module, and the output of the forward module is the second ellipsometric spectrum and the second supplementary spectrum.

[0035] According to an embodiment of the present invention, both the reverse module and the forward module include a plurality of cascaded U-net units. Each U-net unit includes a first convolutional layer, a second convolutional layer, a third convolutional layer, and an upsampling layer; the first convolutional layer includes two residually connected convolutional units, and the second convolutional layer includes two residually connected convolutional units; a convolutional unit of the first convolutional layer, a convolutional unit of the second convolutional layer, the third convolutional layer, the upsampling layer, another convolutional unit of the second convolutional layer, and another convolutional unit of the first convolutional layer are sequentially connected.

[0036] Residual connections are the process of adding up the values ​​of each element in the tensor during the dense block (convolutional unit) connection operation. Alternating connections perform element-wise addition on all nodes that can be connected to the dense block, allowing the tensor information of each node to communicate with each other, preserving the original tensor information while also retaining the depth information obtained through the convolution operation.

[0037] The forward module / reverse module composed of several cascaded U-net units is as follows Figure 4 As shown, the nondestructive testing model formed by the combination of the forward module and the reverse module is as follows Figure 3 As shown. Figure 4 The forward module and the reverse module shown both deal with data regression tasks. For regression tasks, U-net and its variants with encoder-decoder properties are usually the most effective architectures.

[0038] Preferably, the number of U-net units in the reverse module and the forward module is 3. It should be noted that the number of U-net units in the reverse module and the forward module can also be other values.

[0039] The design principle of the nondestructive testing model in this embodiment is described below.

[0040] The forward physical process of ellipsometry is as follows Figure 2 As shown. The principle of ellipsometry is to deduce the properties of materials based on the change in the polarization state of light obliquely incident on the sample surface. The "birefringence" behavior of light obliquely incident on the material has enabled the physical description of the forward process in ellipsometry measurement to be well developed. The mathematical description of this anisotropic behavior was later completed by Fresnel using a set of equations now known as the "Fresnel formula". Afterwards, Airy further proposed a multi-beam interference formula to calculate the reflection coefficient r and transmission coefficient t of the thin film covered surface. Furthermore, the reflectivity R and transmittance T of both p-light and s-light polarizations can be calculated. Since terahertz waves are totally reflected on the metal surface, the transmittance T of the thin film material placed on the metal substrate is 0. Therefore, for an actual terahertz ellipsometer, the film reflectivity R and absorptivity A can be obtained in actual experiments based on the following energy conservation formula:

[0041] R+A+T=1

[0042] Then, using the ellipsometry spectrum (Ψ, Δ) and the complementary spectrum (R, A), a set of film parameters (n, k, d) can be determined analytically and uniquely as shown in the following equation:

[0043]

[0044] Solving the nondestructive testing model is a cyclic and iterative process of updating parameters. Based on deep neural networks, the dataset is full of ellipsometric knowledge learned from offline and online training, which can effectively solve the above formula.

[0045] In operation S2 , the simulated measurement data set is preprocessed to obtain a degraded measurement data set, where samples in the degraded measurement data set include ellipsometric spectra and supplementary spectra.

[0046] According to an embodiment of the present invention, preprocessing the simulated measurement data set specifically includes: performing scaling processing, vector offset processing, and noise addition processing on the simulated measurement data set in sequence. Preferably, the simulated measurement data set is preprocessed using a depth blind degradation model, for example.

[0047] See Figure 5 , shows the experimental setup of the terahertz ellipsometer optical measurement. The terahertz transmitter on the far right radiates terahertz waves with a frequency of 0.3THz through the horn antenna, and then passes through the beam collimating lens to form parallel light that illuminates the 1 / 4 wave plate. It then passes through the focusing lens to form a light spot with a diameter of 5mm on the sample. The light beam is scattered after interacting with the sample, and is collected by the collimating lens and illuminates the 1 / 2 glass slide. Finally, it passes through a focusing lens to collect the light beam to the mouth of the terahertz receiver horn antenna.

[0048] First, by rotating the quarter-wave plate 1° and the half-wave plate 180°, we determine the angle at which the quarter-wave plate converts 0.3 THz linearly polarized light into circularly polarized light. We then adjust the quarter-wave plate's position to ensure that the polarization state of the beam irradiated by the ellipsometer on the sample surface is circularly polarized. Then, with the quarter-wave plate angle fixed, the sample to be measured is placed on the substrate. At this point, we rotate only the half-wave plate 180° to measure the change in polarization state of the beam after it passes through the sample. This generates a simulated measurement data set.

[0049] Since the experimentally measured (Ψ, Δ) and (R, A) data inevitably contain noise, such as light source radiation intensity fluctuations, ellipsometer system errors and detector noise. In order to enable the neural network to acquire ellipsometric knowledge, the non-destructive testing model is first trained offline on a large amount of simulation data. These data may have certain deviations from the ellipsometric data of the real experiment. Therefore, this embodiment uses data degradation and data enhancement methods to improve the robustness of the neural network. The solution (n(t), k(t), d(t)) generated by each iteration is subjected to data degradation operation. First, some random points are generated in the neighborhood of the current solution (n(t), k(t), d(t)), and then the corresponding pseudo-measurement values ​​are calculated using the forward module.

[0050] The specific data degradation method is as follows: 1) Randomly shift the input feature (Ψ, Δ, R, A) to the left or right by a scale not greater than 3; 2) Multiply the input feature (Ψ, Δ, R, A) by a small random scale (1+g*Nr), where g is a random number generated by a standard normal distribution and Nr is a random number not greater than 10 -1 Each data degradation method generates 100 solutions, resulting in a total of 300 solutions. This, combined with the current solution, increases the degradation space of the dataset within the neighborhood of the current solution (n(t), k(t), d(t)). By training the forward module on the processed solution set, the forward module is able to better approximate the true functions F and G within the neighborhood of the current solution.

[0051] During offline training, data processing methods are used to refine the solution (n(t), k(t), d(t)) from the reverse module. The simulation dataset consists of 1980 pairs of (Ψ, Δ, R, A) and (n, k, d). These pairs are based on the theoretical range (n, k) of laboratory-prepared polymer materials in the terahertz band and are mathematically modeled using forward physics functions of F and G in conjunction with MATLAB. d varies between 10 and 900 μm, with a step size of 10 μm for the material thickness.

[0052] Online training reads the pre-trained model obtained by offline training, and then the dataset processing method for online training is the same as the two methods for offline training, but an additional degradation method is introduced here: a random offset current solution is generated by 3×10 -2 The ratio of the simulation data set allocated to offline training and online training is 8:2.

[0053] In operation S3, the ellipsometric spectrum and the supplementary spectrum in the degradation measurement data set are input into the inverse module, and the nondestructive testing model is trained with the goal of minimizing the error between the second ellipsometric spectrum and the second supplementary spectrum output by the forward module and the ellipsometric spectrum and the supplementary spectrum in the degradation measurement data set.

[0054] According to an embodiment of the present invention, the objectives when training a nondestructive testing model are:

[0055] (n * ,k * ,d * )=argmin n,k,d [γ||F(n,k,d)-(ψ,Δ)||2+(1-γ)||G(n,k,d)-(R,A)||2]

[0056] Among them, n * 、k * d * are the optimal n, k, and d, respectively. n, k, and d are the refractive index, extinction coefficient, and thickness of the film, respectively. γ is the weight coefficient. ψ is the amplitude ratio of p-light and s-light. Δ is the phase difference between p-light and s-light. ψ and Δ constitute the ellipsometric spectrum. R is the reflectivity of the film. A is the absorptivity of the film. R and A constitute the complementary spectrum. F() is the first function, G() is the second function, and ||||2 is the Euclidean distance.

[0057] The first function F() is:

[0058]

[0059] The second function G() is:

[0060]

[0061] A=1-R

[0062] Among them, r p is the p light reflection coefficient, r s is the light reflection coefficient.

[0063] In order to bridge the gap between simulation data and real experimental data, the embodiment of the present invention provides a new network iteration method with loop feedback, such as Figure 3 As shown in Figure 2, the reverse module and the forward module work together in a closed loop.

[0064] From the perspective of network operation process, this method takes (Ψ, Δ, R, A) as input and automatically outputs a set of solutions (n, k, d). After proper training, the inverse module is used as a function F -1 and G -1 The forward module is trained as a substitute for the functions F and G, and automatically outputs a set of solutions (Ψ, Δ, R, A) based on the candidate solutions (n(t), k(t), d(t)). Specifically, given a set of data (Ψ(0), Δ(0), R(0), A(0)), a set of solutions (n(0), k(0), d(0)) is calculated through the inverse module. Then, data degradation and data enhancement are performed in a small neighborhood of (n(0), k(0), d(0)) to form a solution set about (n(0), k(0), d(0)). Then, the forward module is executed to simulate the calculation process of F and G, and calculate the ellipsometric data (Ψ(0), Δ(0), R(0), A(0)) reconstructed from the solution set (n(0), k(0), d(0)). Then, the error between the reconstructed ellipsometric data and the true ellipsometric data is compared, and the error value is back-propagated through the two modules to update the trainable weights in the inverse module. The updated inverse module will generate a better solution (n(1), k(1), d(1)). This process is repeated until (n(t), k(t), d(t)) generates (Ψ(t), Δ(t), R(t), A(t)) that is good enough to reconstruct (Ψ(0), Δ(0), R(0), A(0)). At this point, the network performance is considered to be optimal and training is stopped.

[0065] Before the (Ψ, Δ, R, A) data is input into the inverse module, it must be split into two sets: ellipsometric data (Ψ, Δ) and supplementary spectral data (R, A). These data are then multiplied by their respective weighting coefficients and input into the inverse module's SRUM. The forward module also requires weighting coefficients at output. The calculated set (Ψ, Δ, R, A) is assigned its own weighting coefficients, and the error between the calculated set and the true data is calculated and used for error propagation and automatic network gradient optimization.

[0066] Example 2

[0067] A method for nondestructively testing the optical constants and thickness of a multilayer thin film comprises: inputting an ellipsometric spectrum and a supplementary spectrum of the film to be tested into a trained nondestructive testing model obtained by the aforementioned model construction method for nondestructively testing the optical constants and thickness of multilayer thin films, thereby obtaining the refractive index, extinction coefficient, and thickness of the film to be tested (output by an inverse module of the nondestructive testing model). The related technical solutions are the same as those in Example 1 and are not further described here.

[0068] Example 3

[0069] An electronic device includes: a processor; and a memory storing a computer-executable program. When executed by the processor, the program causes the processor to execute the aforementioned model-building method for nondestructively testing optical constants and thicknesses of multilayer thin films, or to execute the aforementioned method for nondestructively testing optical constants and thicknesses of multilayer thin films. The related technical solutions are the same as those in Embodiments 1 and 2 and are not further described here.

[0070] Example 4

[0071] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned model-building method for nondestructively testing the optical constants and thickness of multilayer thin films, or the aforementioned method for nondestructively testing the optical constants and thickness of multilayer thin films. The related technical solutions are the same as those in Examples 1 and 2 and are not further described here.

[0072] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A model construction method for non-destructive detection of optical constants and thickness of multilayer films, characterized in that: include: Constructing a nondestructive testing model; wherein the nondestructive testing model includes an inverse module and a forward module; the input of the inverse module is a first ellipsometric spectrum and a first supplementary spectrum, and the output is the refractive index, extinction coefficient, and thickness of the film; the output of the inverse module serves as the input of the forward module, and the output of the forward module is a second ellipsometric spectrum and a second supplementary spectrum; Preprocessing the simulation measurement data set to obtain a degraded measurement data set, where samples in the degraded measurement data set include ellipsometry spectra and supplementary spectra; Inputting the ellipsometric spectrum and the supplementary spectrum in the degradation measurement data set into the inverse module, and training the nondestructive testing model with the goal of minimizing the errors of the second ellipsometric spectrum and the second supplementary spectrum output by the forward module relative to the ellipsometric spectrum and the supplementary spectrum in the degradation measurement data set; The goals when training the NDT model are: (n * ,k * ,d * )=argmin n,k,d [γ||F(n,k,d)-(ψ,Δ)||2+(1-γ)||G(n,k,d)-(R,A)||2] Among them, n * 、k * d * are the optimal n, k, and d, respectively. n, k, and d are the refractive index, extinction coefficient, and thickness of the film, respectively. γ is the weight coefficient. ψ is the amplitude ratio of p-light and s-light. Δ is the phase difference between p-light and s-light. ψ and Δ constitute the ellipsometric spectrum. R is the reflectivity of the film. A is the absorptivity of the film. R and A constitute the complementary spectrum. F() is the first function, G() is the second function, and || ||2 is the Euclidean distance.

2. The method for constructing a model for nondestructive testing of optical constants and thickness of multilayer thin films according to claim 1, wherein: F() is: G() is: A=1-R Among them, r p is the p light reflection coefficient, r s is the light reflection coefficient.

3. The method for constructing a model for nondestructive testing of optical constants and thickness of multilayer thin films according to any one of claims 1 to 2, characterized in that: The reverse module and the forward module both include a plurality of cascaded U-net units; The U-net unit includes a first convolutional layer, a second convolutional layer, a third convolutional layer and an upsampling layer; The first convolutional layer includes two residual-connected convolutional units, and the second convolutional layer includes two residual-connected convolutional units; A convolution unit of the first convolution layer, a convolution unit of the second convolution layer, the third convolution layer, the upsampling layer, another convolution unit of the second convolution layer, and another convolution unit of the first convolution layer are connected in sequence.

4. The method for constructing a model for nondestructive testing of optical constants and thickness of multilayer thin films according to claim 3, wherein: In the reverse module and the forward module, the number of U-net units is 3.

5. The method for constructing a model for nondestructive testing of optical constants and thickness of multilayer thin films according to claim 1, wherein: The preprocessing of the simulation measurement data set specifically includes: The simulation measurement data set is sequentially subjected to scaling processing, vector offset processing, and noise addition processing.

6. A method for non-destructive testing of optical constants and thickness of multilayer thin films, characterized in that: The ellipsometric spectrum and the supplementary spectrum of the film to be tested are input into the trained nondestructive testing model obtained according to any one of claims 1 to 5 to obtain the refractive index, extinction coefficient and thickness of the film to be tested.

7. An electronic device, characterized in that: include: processor; A memory storing a computer-executable program, wherein when the program is executed by the processor, the processor executes the model building method for non-destructively detecting the optical constants and thickness of a multilayer thin film according to any one of claims 1 to 5, or executes the method for non-destructively detecting the optical constants and thickness of a multilayer thin film according to claim 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the model building method for non-destructively detecting optical constants and thickness of multilayer thin films according to any one of claims 1 to 5 is implemented, or the method for non-destructively detecting optical constants and thickness of multilayer thin films according to claim 6 is implemented.

Images