Method for measuring optical and geometric properties of a thin film material

Abstract

The application discloses a kind of optical properties and geometric characteristics measurement method of thin film material, belong to ellipsometry field, this method includes: using spline model and forward optical property model generates training set;Training neural network model;The preliminary result of the geometric parameter and spline parameter of material obtained by inputting measured optical characterization to neural network is sequentially input spline model and forward optical property model to obtain theoretical optical characterization, the deviation between theoretical and measured optical characterization is used to optimize neural network model, and the predicted value output by it is the final geometric and optical property parameters of the material.The present application does not depend on the experience of the operator, and can realize intelligent characterization;It has strong generalization ability, can solve the problem of inaccurate prediction caused by insufficient training data, and can test samples with different measurement configurations than the training set;It is suitable for measuring various geometric structures;The extraction result is accurate, and the robustness to noise is good.

Description

Technical Field

[0001] This invention belongs to the field of ellipsometry measurement, and more specifically, relates to a method for measuring the optical and geometric properties of thin film materials. Background Technology

[0002] Nanofilms are commonly used structures in semiconductor device manufacturing, and their thickness and optical properties significantly influence device performance. The basic principle of ellipsometric measurement is to measure and analyze the intensity of scattered light after the interaction of incident light with the sample, thereby obtaining the change in the beam's polarization state and extracting the parameters to be determined from the sample. This technique can simultaneously characterize geometric parameters such as the thickness of thin film samples as well as the optical properties of the material. It offers advantages such as low cost and non-destructive operation, and therefore has wide applications in semiconductor measurement.

[0003] Besides measuring instruments, the successful implementation of ellipsometric measurement technology also relies on parameter extraction algorithms. For analyzing new thin film materials, the commonly used algorithm is nonlinear regression, which mainly involves: first, establishing a forward optical property model; second, providing initial values ​​such as the thickness and optical properties of the sample to be measured, and iteratively adjusting the measured parameters to ensure that the theoretical characterization values ​​calculated by the forward optical property model match the measured characterization values. This process is also known as solving the inverse problem of optical scattering. Poor selection of initial values ​​can cause the optimization to get stuck in a local optimum, therefore, technicians need sufficient prior knowledge of the sample, such as the nominal thickness of the sample, and the selection and establishment of the material's optical property model.

[0004] In recent years, the development of machine learning technology has made it possible to directly extract parameters from optical characteristics. To address this issue, existing technologies typically employ the following methods: establishing cascaded neural networks and adding a fitting process containing a forward network to optimize the extraction results; however, the accuracy of the forward network is limited by the data in the training set. Alternatively, U-NET networks can be used to directly extract optical constants; or thin-film neural networks can be used to extract optical constants. However, none of these methods can guarantee that the real and imaginary parts of the predicted optical constants satisfy physical constraints, leading to deviations in the predicted optical constants. Furthermore, machine learning-based extraction methods suffer from poor generalization, meaning that the trained machine learning model can only be applied to specific samples and measurement configurations. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method for measuring the optical and geometric properties of thin film materials. After training a neural network, the method further optimizes the network during the application phase using an optical property spline model and a forward optical property model. Therefore, the accuracy of the neural network's prediction results is unaffected by the training set. Furthermore, since a spline model is used to describe the dielectric function, the real and imaginary parts of the predicted optical properties of the material satisfy physical constraints, thereby improving the accuracy of the measurement results. In addition, during the application phase, the preset measurement conditions corresponding to the forward optical property model can be adjusted according to the actual measurement situation. The neural network is then optimized based on the optical property spline model and the forward optical property model. Therefore, it can be generalized for characterizing different measurement configurations and samples.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a method for measuring the optical and geometric properties of thin film materials is provided, comprising:

[0007] Training phase:

[0008] S1, based on the dielectric functions ε of m different thin film materials respectively j (E)=ε j,1 (E)+iε j,2 (E) and optical property spline model to determine spline parameter b j Within a preset range, the geometric parameters of the thin film sample are randomly selected to obtain multiple sets of geometric parameters x1. ， x2, ..., x n ; each ε j and set x k The theoretical optical characterization is obtained by inputting the forward optical property model corresponding to the measurement conditions during the training phase. To construct a training set; where j∈[1, m], k∈[1, n];

[0009] S2, with As input, the corresponding b j x k As output, the neural network is trained using the training set;

[0010] Application phase:

[0011] S1`, obtain the optical characterization quantity y of the thin film material under the measurement conditions during the application stage. mea And input it into the trained neural network to obtain b pre and x pre ;

[0012] S2`, b pre Input the optical property spline model to obtain ε pre;

[0013] S3`, ε pre and x pre The corresponding theoretical optical characterization quantity y is obtained by inputting the forward optical property model corresponding to the measurement conditions in the application stage. t ; with y mea With y t The trained neural network is optimized with the goal of minimizing the deviation between the two conditions; wherein the measurement conditions in the application stage are the same as or different from the measurement conditions in the training stage.

[0014] S4`, will y mea The input is fed into the optimized neural network to obtain b` pre and x` pre , will b` pre The input is given to the optical property spline model to obtain ε`. t .

[0015] According to a second aspect of the present invention, a system for determining the optical and geometric properties of a thin film material is provided, comprising: a computer-readable storage medium and a processor;

[0016] The computer-readable storage medium is used to store executable instructions;

[0017] The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in the first aspect.

[0018] According to a third aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to perform the method as described in the first aspect.

[0019] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0020] (1) Compared with traditional extraction methods, the method proposed in this invention can get rid of the dependence on the engineering experience of technicians and realize the intelligentization of the measurement process.

[0021] (2) During the optimization of the neural network model by the spline model and the forward optical property model, the measurement configuration can be changed, so it can be generalized to characterize samples under different measurement configurations.

[0022] (3) In this method, some outputs of the neural network are ignored, and changing the settings of the spline model and the positive optical property model will not affect the entire extraction process. Therefore, it can be flexibly applied to the testing of samples with various geometric structures.

[0023] (4) Compared with existing representation methods, the accuracy of neural network prediction results in the method provided by this invention is not affected by insufficient dataset.

[0024] (5) Compared with existing characterization methods, the introduction of spline model can ensure that the extracted optical properties of the material have physical meaning. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for measuring the optical and geometric properties of thin film materials provided in an embodiment of the present invention;

[0026] Figure 2 These are geometric feature maps of the samples used to construct the training set and the first test sample in this embodiment of the invention.

[0027] Figure 3 This is a geometric feature diagram of test sample two in this embodiment of the invention;

[0028] Figure 4 This is a geometric feature diagram of test sample three in this embodiment of the invention;

[0029] Figure 5 This is a comparison chart of the characterization results of test sample one in the embodiments of the present invention;

[0030] Figure 6 This is a comparison chart of the characterization results of test sample two in this embodiment of the invention;

[0031] Figure 7 This is a comparison chart of the characterization results of test sample three in this embodiment of the invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be 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 illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0033] This invention provides a method for measuring the optical and geometric properties of thin film materials, including:

[0034] Training phase:

[0035] S1, based on the dielectric functions ε of m different thin film materials respectively j (E)=ε j,1 (E)+iε j,2 (E) and optical property spline model to determine spline parameter b jDetermine the specific geometric structure of the thin film sample: Randomly select geometric parameters of the thin film sample within a preset range to obtain multiple sets of geometric parameters x1, x2, ..., x n ; each ε j and set x k The theoretical optical characterization is obtained by inputting the forward optical property model corresponding to the measurement conditions during the training phase. To construct a training set; wherein, the training set consists of and with The corresponding b j x k constitute.

[0036] Specifically, such as Figure 1 As shown, firstly, a spline model of material optical properties with fast computation speed and physical meaning is established, which can calculate the optical property curves of given nodes and coefficients.

[0037] The optical property spline model can reconstruct the curves of the real and imaginary parts of the material's dielectric function from a set of coefficients under a certain node configuration. Furthermore, the formula for calculating the real part can be derived from the formula for calculating the imaginary part using the Kramers–Kronig consistency relation (and vice versa). B-spline models are commonly used. Node configuration includes setting the position and number of nodes.

[0038] Preferably, an optical property spline model is established based on a B-spline model;

[0039] Alternatively, an optical property spline model can be established based on the B-spline model and the Pole oscillator model.

[0040] Furthermore, in this embodiment, an optical property spline model based on a cubic B-spline and a Pole oscillator model is established, wherein the spline parameters include B-spline coefficients and oscillator parameters, denoted as b = [b1, b2, ..., b M ] T In the formula, M represents the total number of spline parameters. Based on the given spline parameters, the real part ε1 and the imaginary part ε2 of the dielectric function ε=ε1+iε2 of the material can be calculated, each being a 1×N vector, where N is the number of preset measurement wavelengths.

[0041] The optical property spline model established based on the cubic B-spline and the Pole oscillator model is as follows:

[0042]

[0043]

[0044] Where E is the photon energy at the measurement wavelength. and φ i 3(E) are the cubic basis functions used to calculate the imaginary and real parts of the dielectric function, respectively. The basis functions can be obtained from the following equation:

[0045]

[0046]

[0047]

[0048] Among them, t i Let k be a node of the spline, k∈[1,3].

[0049] It is worth noting that the basis functions of the real part in the optical property spline model are derived from the basis functions of the imaginary part by the Kramers-Kronig consistency condition, which ensures that ε1 and ε2 maintain Kramers-Kronig consistency. Furthermore, matrix operations can be used instead of traditional recursive algorithms in the programming implementation, allowing for the simultaneous calculation of different basis functions across the entire wavelength range, thus enabling rapid calculation of the dielectric function curve.

[0050] Based on the dielectric function ε(E)=ε1(E)+iε2(E) and the optical property spline model of each thin film sample of different materials, least squares fitting is performed to obtain the spline parameters b=(b1,…,b m ), where m is the number of spline parameters.

[0051] The thin film samples of the above-mentioned different materials may include at least one of semiconductors (such as silicon, germanium, etc.), oxides (such as titanium oxide, zinc oxide, etc.), and metals (such as gold, copper, etc.).

[0052] Secondly, a rapid forward optical property model is established, which can calculate the optical characterization quantities corresponding to the geometric features and material optical properties of a given sample under a preset measurement configuration.

[0053] Then, spline parameters of optical property curves of various materials are obtained according to spline model. Using the established forward optical property model and the preset measurement configuration (i.e. preset measurement conditions: application stage measurement conditions), the geometric parameters of the sample are randomly selected within the specified parameter range to calculate the corresponding theoretical optical characterization, thereby obtaining a training dataset containing optical characterization, spline parameters and geometric parameters.

[0054] Preferably, a forward optical characteristic model can be established based on the thin film transfer matrix algorithm. In terms of programming method, matrix operations are used to replace the loop recursive algorithm, which can simultaneously calculate the theoretical optical characterization quantities under different measurement configurations (i.e., preset measurement conditions, such as incident angle, film material, etc.) across the entire band, thereby achieving the purpose of rapid calculation.

[0055] Furthermore, the forward optical property model can also be established based on rigorous coupled-wave analysis, boundary element method, or finite-time difference method; and it can be programmed using the idea of ​​matrix operations, which has an extremely fast calculation speed.

[0056] Preferably, the geometric parameter is at least one of thickness, roughness, and non-uniformity.

[0057] Preferably, the optical characterization quantity is at least one of reflectivity, transmittance, ellipticity parameter, and Mueller matrix.

[0058] In this embodiment, the optical characterization quantities are selected as the (N,C,S) spectrum at an incident angle of 65° and the transmittance spectrum at an incident angle of 0°, denoted as y=[y1,y2,...,y n×4 ] T In the formula, n is the number of measurement wavelengths; in the preset measurement conditions, the measurement wavelength is set to 1.26 to 4.13 eV, the wavelength interval is 0.01 eV, and the substrate is 0.6 mm fused silica glass; Figure 2 The geometric features of the thin film sample structure are given, denoted as x = [x1, x2, x3]. T , representing the thickness, roughness and non-uniformity of the geometric parameters to be measured, respectively. The spline model sets equally spaced nodes within the measurement band with an interval of 0.07 eV, and additional nodes are set at positions of 0.66, 0.86, 1.06, 4.63, 5.13, 6.13 and 8.13 eV outside the measurement band to account for out-of-band absorption.

[0059] In the training data constructed in this embodiment, the geometric parameters to be tested are set as film thickness, roughness layer thickness and non-uniformity, and the substrate is set as a single 0.6 mm thick fused silica film. The preset ranges include: the parameter range of film thickness is 20-200 nm, the parameter range of roughness layer thickness is 0-10 nm, and the parameter range of non-uniformity is -100 to 100%.

[0060] By fitting the dielectric functions of m different thin film materials using a spline model, multiple sets of spline parameters are obtained. Based on the random combination of known dielectric functions and geometric parameters, multiple sets of theoretical optical characterizations can be generated through a forward optical property model. The training set is composed of the optical characterizations and their corresponding spline and geometric parameters.

[0061] S2, train the neural network using the training set, to As input, and with The corresponding b j x k As output;

[0062] Specifically, a neural network model is constructed, with optical representations as inputs, spline parameters and geometric parameters as outputs, and the deviation between the network's predictions and theoretical values ​​as the loss function. This loss function is then used to train the neural network model.

[0063] Preferably, the neural network is a fully connected neural network or a convolutional neural network.

[0064] Specifically, neural network models are highly flexible and can ignore irrelevant output values ​​to achieve measurements of various geometries (such as with or without roughness).

[0065] For example, constructing a residual convolutional neural network to incorporate optical representations from the training set. As input to this neural network, the spline parameter b j and geometric parameters x k As output, the network prediction results (b) j ,x k ) pre Compared with the theoretical value (b) j ,x k ) t The deviation between the two is used as the loss function, where the deviation is calculated as mean squared error, and the constructed neural network model is trained by reducing this loss function.

[0066] Application phase:

[0067] S1`, obtain the optical characterization quantity y of the thin film material under the measurement conditions during the application stage. mea And input it into the trained neural network to obtain b pre and x pre ;

[0068] Specifically, based on the neural network trained by S2, the optical characterization quantity y is measured. mea As input, preliminary results of its corresponding geometric parameters and spline parameters can be obtained. pre and x pre ;

[0069] Preferably, in step S1', the optical characterization of the thin film material under test is obtained by using an ellipsometer under the measurement conditions of the application stage.

[0070] Preferably, the measurement conditions in both the application phase and the training phase include: incident angle, measurement wavelength, substrate material, and thickness.

[0071] S2`, b pre and x pre Inputting the optical property spline model yields y t ; with y meaWith y t The trained neural network is optimized with the goal of minimizing the deviation between the two sides.

[0072] Specifically, based on the spline model and the forward optical property model, the theoretical optical characterization quantity y of the preliminary results can be obtained. t The deviation between the theoretical and measured optical characterizations is calculated, and this deviation is used to optimize the neural network model based on the gradient backpropagation algorithm.

[0073] Preferably, in step S2', the deviation is the mean square error or the mean absolute error. Optimization conditions include the maximum number of iterations, the minimum deviation threshold, etc.

[0074] Based on the spline parameters b given by the neural network model pre The dielectric function ε of the material is calculated using a spline model. pre Based on the geometric parameters x given by the neural network model pre The dielectric function ε calculated by spline model pre The theoretical optical characterization quantity y can be calculated using the forward optical property model. t The measurement conditions set in the spline model and the forward optical property model can be modified according to the specific experimental measurement conditions of the sample without affecting the subsequent calculation process. Examples include changing the incident angle, measurement band, substrate material and thickness, and reducing the number of spline model nodes. In other words, the measurement conditions in the application phase can be the same as or different from those in the training phase.

[0075] In the optimization process of a neural network model, if certain outputs of the neural network are not used, they can be directly omitted during the calculation process. For example, if roughness thickness, non-uniformity, or the number of spline model nodes are not used, this will not affect the optimization of the neural network model.

[0076] S3`, ε pre and x pre The corresponding theoretical optical characterization quantity y is obtained by inputting the forward optical property model corresponding to the measurement conditions in the application stage. t ; with y mea With y t The trained neural network is optimized with the goal of minimizing the deviation between the two conditions; wherein the measurement conditions in the application stage are the same as or different from the measurement conditions in the training stage.

[0077] Specifically, when the optimization conditions are met, the output value of the neural network with the smallest deviation is the final geometric and spline parameters of the sample. By further utilizing the spline model and the forward optical property model, the geometric features and optical properties of the thin film material under test can be characterized.

[0078] For example, the optimization conditions can be set as follows: a maximum number of iterations of 1000 and a deviation threshold of 10. -6 If the optimized neural network model outputs the corresponding theoretical optical characterization y... t With measurement of optical characterization quantity y mea If the deviation between the parameters exceeds a threshold, optimization continues until the stopping condition is met. After optimization, the neural network with the smallest deviation is the optimal model. Based on its output value, the geometric parameters x and spline parameters b of the sample can be determined. Then, the dielectric function ε of the sample can be obtained using the spline model, thereby characterizing the sample.

[0079] Understandably, during the application phase, the parameters related to measurement condition configuration in the forward optical characteristic model (such as incident angle, measurement band, spline node distribution, substrate material and thickness, etc.) can be modified according to the new sample to be tested. That is, the measurement conditions in the application phase can be the same as or different from those in the training phase.

[0080] In summary, the method provided by this invention first trains a neural network based on a pre-generated dataset, enabling it to predict the geometric property parameters corresponding to a given optical characterization quantity. When characterizing a new sample, the neural network is optimized using a spline model and a forward optical property model, and the result obtained from the optimal neural network is the final parameter. Through this invention, the accuracy of the neural network is unaffected by the dataset, and the real and imaginary parts of the predicted optical properties of the material satisfy physical constraints, allowing it to be generalized for characterizing different measurement configurations and samples.

[0081] The following section presents the optical and geometric characterization results of three different thin film samples, using ellipsometer-based measurements of nanofilms as an example:

[0082] In this embodiment, the optical characterization quantities are selected as the (N,C,S) spectrum at an incident angle of 65° and the transmittance spectrum at an incident angle of 0°, denoted as , where n is the number of measurement wavelengths; in the preset measurement conditions, the measurement wavelengths are set to 1.26 to 4.13 eV, with a wavelength interval of 0.01 eV; Figure 2 The geometric features of the thin film sample structure are given, denoted as x = [x1, x2, x3]. T , representing the thickness, roughness and non-uniformity of the geometric parameters to be measured, respectively. The spline model sets equally spaced nodes within the measurement band with an interval of 0.07 eV, and additional nodes are set at positions of 0.66, 0.86, 1.06, 4.63, 5.13, 6.13 and 8.13 eV outside the measurement band to account for out-of-band absorption.

[0083] Sample 1 is a non-uniform zinc oxide film on a fused silica substrate, the geometry of which is as follows: Figure 2As shown. The actual measured incident angle of the (N,C,S) spectrum of this sample was 65.06°, and the incident angle of the transmittance spectrum was 0°. The measurement wavelengths were set from 1.26 to 4.13 eV with a wavelength interval of 0.01 eV. The spline model had equally spaced nodes within the measurement band with an interval of 0.07 eV, and additional nodes were set at positions of 0.66, 0.86, 1.06, 4.63, 5.13, 6.13, and 8.13 eV outside the measurement band. The thickness of the fused silica substrate was 0.6 mm. Manual analysis revealed that the film thickness of this sample was 130.84 nm, the roughness was 3.51 nm, and the non-uniformity was 4.23%. Figure 5 The results of the characterization method provided in this invention are presented. The film thickness of the sample is 130.14 nm, the roughness is 3.6 nm, and the non-uniformity is 4.3%. The dielectric function obtained by the method provided in this invention and by manual analysis is consistent.

[0084] Sample 2 has a substrate of silicon and silicon dioxide films, and the film to be tested is titanium dioxide, with the following geometry: Figure 3 As shown. The actual measured incident angle of the (N,C,S) spectrum of this sample was 65.06°, the transmittance spectrum was set to zero, the measurement wavelength was set to 1.26 to 4.13 eV with a wavelength interval of 0.01 eV, the spline model had equally spaced nodes within the measurement band with an interval of 0.07 eV, and additional nodes were set at positions of 0.66, 0.86, 1.06, 4.63, 5.13, 6.13, and 8.13 eV outside the measurement band. The silica thickness was 201.76 nm. Manual analysis showed that the film thickness of this sample was 76.79 nm and the roughness was 2.54 nm. Figure 6 The results of the characterization method provided in this invention are presented. The film thickness of the sample is 75.05 nm, and the roughness is 4.2 nm. The dielectric functions obtained by the two methods are consistent.

[0085] The gold thin film on the Schottky glass substrate of sample three has the following geometry: Figure 4 As shown. The actual measured incident angle of the (N,C,S) spectrum of this sample was 65.06°, and the incident angle of the transmittance spectrum was 0°. The measurement wavelengths were set from 1.26 to 4.13 eV with a wavelength interval of 0.01 eV. The spline model had equally spaced nodes within the measurement band with an interval of 0.07 eV, and additional nodes were set at positions of 0.66, 0.86, 1.06, 4.63, 5.13, 6.13, and 8.13 eV outside the measurement band. The Schottky glass substrate thickness was 0.5 mm. Manual analysis determined the film thickness of this sample to be 24.05 nm. Figure 7 The results of the characterization method provided in this invention are presented, showing a film thickness of 24.06 nm for the sample. The dielectric functions obtained by the two methods are consistent.

[0086] This invention provides a system for determining the optical and geometric properties of thin film materials, comprising: a computer-readable storage medium and a processor;

[0087] The computer-readable storage medium is used to store executable instructions;

[0088] The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any of the above embodiments.

[0089] This invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions, which are used to cause a processor to perform the method described in any of the above embodiments.

[0090] Those skilled in the art will readily understand 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 within the scope of protection of the present invention.

Claims

1. A method for measuring the optical and geometric properties of a thin film material, characterized in that, include: Training phase: S1, based on the dielectric functions of m different thin film materials respectively And optical property spline model to determine spline parameter b j Within a preset range, the geometric parameters of the thin film sample are randomly selected to obtain multiple sets of geometric parameters. ; each ε j and sets The theoretical optical characterization is obtained by inputting the forward optical property model corresponding to the measurement conditions during the training phase. , to construct a training set; where j∈[1,m], k∈[1,n]; S2, with As input, the corresponding b j , As output, the neural network is trained using the training set; Application phase: S1`, obtain the optical characterization quantity y of the thin film material under the measurement conditions during the application stage. mea The results are then input into a trained neural network to obtain preliminary results of the spline parameters and geometric parameters of the thin film material under test under the measurement conditions during the application stage. pre and x pre ; S2`, b pre By inputting the optical property spline model, preliminary results ε of the dielectric function of the thin film material under test under the measurement conditions in the application stage are obtained. pre ; S3`, ε pre and x pre The corresponding theoretical optical characterization quantity y is obtained by inputting the forward optical property model corresponding to the measurement conditions in the application stage. t ; with y mea With y t The trained neural network is optimized with the goal of minimizing the deviation between the two conditions; wherein the measurement conditions in the application stage are the same as or different from the measurement conditions in the training stage. S4`, will y mea The input is fed into the optimized neural network to obtain b` pre and x` pre , will b` pre The final result ε` of the dielectric function of the thin film material under test is obtained by inputting the optical property spline model. t .

2. The method as described in claim 1, characterized in that, In step S1, an optical property spline model is established based on the B-spline model; Alternatively, an optical property spline model can be established based on the B-spline model and the Pole oscillator model.

3. The method as described in claim 2, characterized in that, When using a cubic B-spline model, the optical property spline model established based on the B-spline model and the Pole oscillator model is as follows: ； Where E is the photon energy at the measurement wavelength. and The cubic basis functions and spline parameters are used to calculate the imaginary and real parts of the dielectric function, respectively. M is the number of spline parameters.

4. The method as described in claim 1, characterized in that, The forward optical characteristic model is established based on the thin film transfer matrix method, rigorous coupled-wave analysis, boundary element method or finite-time difference method, and the corresponding measurement conditions; the measurement conditions are the measurement conditions in the application stage or the measurement conditions in the training stage.

5. The method as described in claim 1, characterized in that, In step S1', the optical characterization quantities of the thin film material under test are obtained under the measurement conditions in the application stage using an ellipsometer; In step S2', the deviation is the mean square error or the mean absolute error.

6. The method as described in claim 1, characterized in that, The optical characterization quantity is at least one of reflectivity, transmittance, ellipticity parameter, and Mueller matrix; The geometric parameter is at least one of thickness, roughness, and non-uniformity.

7. The method as described in claim 1 or 4, characterized in that, The measurement conditions for the application phase include: incident angle, measurement wavelength, substrate material and thickness.

8. The method as described in claim 1, characterized in that, The neural network is either a fully connected neural network or a convolutional neural network.

9. A system for determining the optical and geometric properties of a thin film material, characterized in that, include: Computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a processor to perform the method as described in any one of claims 1-8.

Images