An artificial intelligence anti-disturbance lithography objective wave aberration measurement method

Abstract

The application discloses an artificial intelligence anti-disturbance lithography objective wave aberration measurement method, and relates to the field of lithography objective wave aberration measurement algorithms. In the application, a U-shaped neural network structure based on physical information is used in a general U-NET neural network model, the overall encoder-decoder structure is retained, and the pooling layer is removed. Compared with the traditional four-step phase shifting (or five-step phase shifting), the application has the following innovative points: (1) on the hardware, the piezoelectric phase shifter only needs to be moved once, and only two-step phase shifting is needed. Compared with the traditional phase shifting wave aberration measurement system, the application has faster measurement speed. (2) on the hardware, the traditional wave aberration measurement system needs an ultrahigh-precision piezoelectric phase shifter, while the application only needs a low-precision standard (the precision is better than 50 nm) piezoelectric phase shifter. (3) on the algorithm, the application can accurately measure the lithography objective wave aberration from two interference patterns. (4) on the wave aberration reconstruction precision, the application has higher precision due to the reduction of environmental interference.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of lithography exposure technology, and particularly relates to an artificial intelligence anti-disturbance lithography objective wave aberration measurement method. BACKGROUND

[0002] Lithography exposure is a core subsystem in a lithography machine, and a high numerical aperture (NA) exposure system can significantly improve the exposure resolution of the lithography machine and improve the process of a semiconductor process. At present, high-NA objectives are mostly used in lithography exposure, and wave aberration detection, such as a five-step phase-shifting method (or other multi-step phase-shifting methods), is complex and is easily disturbed by external vibration, air flow disturbance, piezoelectric phase shifter error and other factors (J. H. Bruning, D. R. Herriott, J. E. Gallagher, D. P. Rosenfeld, A. D. White, and D. J. Brangaccio, "Digital Wavefront Measuring Interferometer for Testing Optical Surfaces and Lenses," Appl. Opt. 13, 2693-2703 (1974)).

[0003] In order to solve this problem, the document (Feng P, Li ZL, Wang ZC, et al. Polarization phase shift point diffraction interference wave aberration detection technology research[J]. China Laser, 2022, 49(21): 90-97.) proposes to use a polarization array technology to simultaneously measure four-step phase shifting, so as to realize the anti-interference capability. The technology uses a short coherence length light source and a single mode polarization maintaining optical fiber to generate two point sources, and the two point sources can output orthogonal linearly polarized light; a quarter wave plate is added in the optical path, and a micro polarization array camera is used to realize spatial synchronous phase shifting based on a single interference image. By adjusting the light intensity ratio of the two beams through an attenuator, the contrast of the interference fringes can be adjusted, but a polarization device with high polarization accuracy is required, and it is not conducive to wave aberration measurement of the working wavelength.

[0004] From the perspective of mathematical model, wave aberration quantitative accurate measurement is an underdetermined (or overdetermined) nonlinear equation system inverse problem. From a more general perspective, researchers have proposed a solution based on artificial intelligence deep learning for these underdetermined (or overdetermined) nonlinear equation system inverse problems in optical imaging, machine vision and holography. Through the enhancement of deep learning, these underdetermined (or overdetermined) nonlinear equations are solved in a faster speed, higher accuracy and less data manner. In lithography objective wave aberration measurement, disturbance error sources make such nonlinear equation system model more complex, increase the difficulty of solving, and lead to a bottleneck in the accuracy of lithography machine wave aberration measurement. SUMMARY

[0005] The present application aims at the above-mentioned deficiencies in the prior art and provides an artificial intelligence anti-disturbance lithography objective wave aberration measurement method.

[0006] The present application aims at the above-mentioned deficiencies in the prior art and provides an artificial intelligence anti-disturbance lithography objective wave aberration measurement method.

[0007] A laser for emitting measuring light;

[0008] A beam splitter connected to the laser through an incident optical fiber, and also connected to a test optical fiber and a reference optical fiber, for splitting the light emitted by the laser into two beams, one entering the test optical fiber as test light and the other entering the reference optical fiber as reference light;

[0009] A piezoelectric phase shifter arranged in the reference optical fiber for phase-shifting the reference light;

[0010] A pinhole aperture having a small hole with a small aperture and a large hole with a large aperture, the small hole being circular and the large hole being a regular polygon or circular;

[0011] An image sensor for detecting and recording an interference pattern;

[0012] The artificial intelligence anti-disturbance lithography objective wave aberration measurement method comprises the following steps:

[0013] (1) The end faces of the test optical fiber and the reference optical fiber are leveled and fixed on the mask surface of the to-be-measured lithography objective, and the positions fixed on the mask surface are symmetrical about the optical axis of the to-be-measured lithography objective; the pinhole aperture is fixed on the chip surface of the to-be-measured lithography objective, and the small hole and the end face of the reference optical fiber are in a conjugate imaging relationship, the large hole and the test optical fiber are in a conjugate imaging relationship, the light emitted by the small hole and the light emitted by the large hole form an interference pattern on the image sensor, which is recorded as a first interference pattern by the image sensor; the piezoelectric phase shifter is controlled to shift by π / 2, which is recorded as a second interference pattern;

[0014] (2) A PIUN network is trained and obtained, which is a U-shaped neural network structure based on physical information;

[0015] (3) The first interference pattern is input into the PIUN network to generate two sequences of interference patterns {I1, I2, I3, I4, I5} and {I1, I'2, I'3, I'4, I'5}, in which I1 represents the interference pattern when the phase of the piezoelectric phase shifter is 0, and the phase difference value of the piezoelectric phase shifter corresponding to the subsequent interference pattern and the previous interference pattern is 2π / 5;

[0016] I2 in the sequence is taken out and structure similarity operation is performed on the second interferogram, if the similarity exceeds 0.7, then {I1, I2, I3, I4, I5} is taken as interferogram data and input into formula 1; if the similarity is lower than 0.7, then {I1, I'2, I'3, I'4, I'5} is taken as interferogram data and input into formula 2;

[0017] Formula 1 is:

[0018] ,

[0019] Wherein, Phi is wave aberration;

[0020] Formula 2 is:

[0021] ,

[0022] Wherein, PhyLoss is physical loss function,

[0023] , , , represents the derivative of the interferogram along the horizontal direction of the picture, , , , represents the derivative of the interferogram along the vertical direction of the picture.

[0024] Preferably, in step (2), a series of interferograms and corresponding wave aberrations are measured, the interferograms are denoted as {I k1 , I k2 , I k3 , I k4 , I k5}, and the wave aberrations corresponding to the interferograms are Phi k , wherein k takes a value from 1 to the total number of the series of interferograms;

[0025] A deep learning data set {I k1 , I k2 ; Phi k} is constructed, the deep learning data set is divided into training data and verification data, the training data is input into an initial PIUN network for training, and the verification data is used for testing until the PIUN network converges, and when the loss function of the PIUN network converges to 10 -7 times, the trained PIUN network is obtained.

[0026] Preferably, the measurement is performed in a high-vacuum and high-stability point diffraction wave aberration measurement system, the high vacuum refers to a vacuum degree ≤ 1 × 10 -7Pa, high stability refers to VC-F level shock absorption environment.

[0027] Preferably, the PIUN network is composed of an image input layer, an encoder, a decoder and an image output layer, the image input layer is one, the image output layer is five, each image output layer is in parallel relationship, and each image output layer corresponds to an independent encoder and a decoder.

[0028] More preferably, each encoder is composed of 10 encoding modules, and each decoder is composed of 10 decoding modules,

[0029] Each encoding module is composed of a 3*3 convolution module, a normalization module and a Leaky Relu activation module;

[0030] Each decoding module is composed of a 3*3 deconvolution module, a normalization module and a Leaky Relu activation module.

[0031] Preferably, each of the test optical fiber and the reference optical fiber is provided with an energy adjuster for adjusting the energy of light, so as to cope with different transmittance of the measured lithography objective lens, and the interference fringe contrast is the highest.

[0032] More preferably, the energy ratio of the reference light to the test light after the beam splitter is 1:7-11, and most preferably 1:9. The energy ratio is realized by adjusting two energy adjusters.

[0033] Preferably, the phase shift accuracy of the piezoelectric phase shifter is ≤50nm. In this application, only a piezoelectric phase shifter with low accuracy standard is needed, which reduces the requirement for the piezoelectric phase shifter in hardware.

[0034] Preferably, the diameter of the small hole is 50%-95% of the wavelength value of the light emitted by the laser. For example, the wavelength of the laser is 633nm, and the diameter of the small hole can be 500nm. When the large hole is a regular polygon, the inscribed circle area of the regular polygon is greater than or equal to the pupil area of the to-be-measured lithography objective lens, and when the large hole is a circle, the circular area is greater than or equal to the pupil area of the to-be-measured lithography objective lens.

[0035] In the algorithm of wavefront aberration measurement of lithography objective, the application innovatively proposes a U-shaped neural network structure based on physical information, which is Physics-informed U-Net in English, abbreviated as PIUN network. In the general U-NET neural network model, the overall encoder-decoder structure is retained, and the pooling layer is removed. We only use convolution and deconvolution operations to adjust the feature dimension of the image. In order to better adapt to computers with different performance, we use "batch" normalization layer in the encoder instead of "layer" normalization layer. Innovatively, in the application, the input and output of the U-NET neural network are not "one-to-one"; but "one-to-five". Innovatively, the role of the network model not only includes predicting the phase shift interferogram, but also includes denoising and equalizing light intensity.

[0036] Compared with the traditional four-step phase shift (or five-step phase shift), the application includes the following innovations: (1) On the hardware, the piezoelectric phase shifter only needs to move once, and only two-step phase shift needs to be obtained. Compared with the traditional phase shift wavefront aberration measurement system, the application has faster measurement speed. (2) On the hardware, the traditional wavefront aberration measurement system needs an ultra-high precision piezoelectric phase shifter, while the application only needs a low-precision standard (precision better than 50nm) piezoelectric phase shifter. (3) On the algorithm, the application can accurately measure the wavefront aberration of the lithography objective from two interferograms. (4) On the wavefront aberration reconstruction precision, the application has higher precision due to the reduction of environmental interference. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is a structural schematic diagram of the lithography objective wavefront aberration measurement device of the application.

[0038] Figure 2 It is a structural schematic diagram of the aperture diaphragm.

[0039] Figure 3 It is a schematic diagram of the artificial intelligence anti-disturbance wavefront aberration measurement method of the application.

[0040] Figure 4 It is the deep learning network architecture of the application.

[0041] Figure 5 It is the structure of the PIUN network.

[0042] Figure 6 It is the structure of the encoder.

[0043] Figure 7 It is the structure of the decoder.

[0044] Figure 8 It is the structure of each encoding module.

[0045] Figure 9 It is the structure of each decoding module.

[0046] Figure 10 The physical function prior loss function of the application.

[0047] Figure 11 The performance improvement experimental data test result of the application.

[0048] The figure mark: laser 1, incident optical fiber 2, beam splitter 3, test optical fiber 4, energy regulator 5, reference optical fiber 6, energy regulator 7, piezoelectric phase shifter 8, photolithography objective 9, mask surface 10, pinhole aperture 11, small hole 12, large hole 13, image sensor 14. DETAILED DESCRIPTION

[0049] As Figure 1 The photolithography objective wave aberration measuring device used in the photolithography objective wave aberration measuring method of the application, including laser 1, incident optical fiber 2, beam splitter 3, test optical fiber 4, energy regulator 5, reference optical fiber 6, energy regulator 7, piezoelectric phase shifter 8, pinhole aperture 11, image sensor 14.

[0050] The laser 1 is used to emit the light for measurement. The beam splitter 3 is connected with the laser 1 through the incident optical fiber 2, and the beam splitter 3 is also connected with the test optical fiber 4 and the reference optical fiber 6. The beam splitter 3 is used to divide the light emitted by the laser 1 into two beams, one of which enters the test optical fiber 4 as test light, and the other of which enters the reference optical fiber 6 as reference light.

[0051] The energy regulator 5 is arranged in the test optical fiber 4, and the energy regulator 7 is arranged in the reference optical fiber 6. The energy regulator is used to cope with the different transmittance of the measured photolithography objective, so that the interference fringe contrast is the highest. After the beam splitter 3 divides, the energy ratio of the reference light to the test light is 1:7~11, and the most preferred is 1:9. The energy ratio is realized through the adjustment of the two energy regulators.

[0052] The piezoelectric phase shifter 8 is arranged in the reference optical fiber 6 and is used to phase shift the reference light. In a preferred embodiment, the phase shift accuracy of the piezoelectric phase shifter is ≤50nm. In the present application, only a piezoelectric phase shifter with low accuracy standard is required, which reduces the requirement for the piezoelectric phase shifter in hardware. Of course, if a piezoelectric phase shifter with higher phase shift accuracy is used, it is also applicable, but the hardware investment needs to be increased.

[0053] As Figure 2As shown, the aperture stop 11 has a small aperture 12 with a small aperture diameter and a large aperture 13 with a large aperture diameter. The small aperture 12 is circular, and the large aperture 13 is a regular polygon or circular. The diameter of the small aperture 12 is 50% to 95% of the wavelength of the light emitted by the laser 1. When the large aperture 13 is a regular polygon, the inscribed circle of the regular polygon has an area greater than or equal to the pupil area of the lithography objective 9 to be measured. When the large aperture 13 is circular, the circular area is greater than or equal to the pupil area of the lithography objective 9 to be measured. In an embodiment, the small aperture 12 is circular with a diameter of 500 nm, and the large aperture 13 is square with a side length of 10 mm.

[0054] The image sensor 14 is used to detect and record the interference pattern.

[0055] The lithography objective 9 to be measured has a mask surface 10 on one side and a chip surface on the opposite side.

[0056] During detection, the end faces of the test optical fiber 4 and the reference optical fiber 6 are flush and fixed on the mask surface 10 of the lithography objective 9 to be measured, and the positions fixed on the mask surface 10 are symmetrical about the optical axis of the lithography objective 9 to be measured. The aperture stop 11 is fixed on the chip surface of the lithography objective 9 to be measured, and the small aperture 12 and the end face of the reference optical fiber 6 are in a conjugate imaging relationship, the large aperture 13 and the test optical fiber 4 are in a conjugate imaging relationship, and the light emitted by the small aperture 12 and the light emitted by the large aperture 13 form an interference pattern on the image sensor 14.

[0057] An artificial intelligence anti-disturbance lithography objective wave aberration measurement method, comprising the following steps:

[0058] (1) The end faces of the test optical fiber and the reference optical fiber are flush and fixed on the mask surface of the lithography objective to be measured, and the positions fixed on the mask surface are symmetrical about the optical axis of the lithography objective to be measured. The aperture stop is fixed on the chip surface of the lithography objective to be measured, and the small aperture and the end face of the reference optical fiber are in a conjugate imaging relationship, the large aperture and the test optical fiber are in a conjugate imaging relationship, and the light emitted by the small aperture and the light emitted by the large aperture form an interference pattern on the image sensor, which is recorded as a first interference pattern; the piezoelectric phase shifter is controlled to shift by π / 2, and a second interference pattern is recorded.

[0059] (2) A PIUN network is trained and obtained, and the PIUN network is a U-shaped neural network structure based on physical information.

[0060] In step (2), a series of interference patterns and corresponding wave aberrations are measured, and the interference patterns are denoted as {I k1 , I k2 , I k3 , I k4 ,I k5}, and the wave aberrations corresponding to the interference patterns are Phi k , where k takes a value from 1 to the total number of the series of interference patterns.

[0061] Constructing deep learning data set k1 , I k2 ;Phi k}, the deep learning data set is divided into training data and validation data, the training data is input into the initial PIUN network for training, and the validation data is tested until the PIUN network converges, when the loss function of the PIUN network converges to 10 -7 , that is, the trained PIUN network is obtained.

[0062] When measuring, it is carried out in a high-vacuum and high-stability point diffraction wave aberration measurement system, the high vacuum refers to a vacuum degree ≤1×10 -7 Pa, and the high stability refers to a VC-F level shock absorption environment.

[0063] The PIUN network is composed of an image input layer, an encoder, a decoder and an image output layer, the image input layer is 1, the image output layer is 5, each image output layer is in parallel relationship, and each image output layer corresponds to an independent encoder and decoder.

[0064] Each encoder is composed of 10 encoding modules, each decoder is composed of 10 decoding modules, each encoding module is composed of a 3×3 convolution module, a normalization module and a Leaky Relu activation module; each decoding module is composed of a 3×3 deconvolution module, a normalization module and a Leaky Relu activation module.

[0065] (3) input the first interference figure into the PIUN network to generate two sequences of interference figures {I1, I2, I3, I4, I5} and {I1, I’2, I’3, I’4, I’5}, in which 5 interference figures of each sequence, I1 represents the interference figure when the phase of the piezoelectric phase shifter is 0, and the phase difference value of the piezoelectric phase shifter corresponding to the subsequent interference figure and the previous interference figure is 2π / 5;

[0066] Take out I2 in the sequence and perform structure similarity operation with the second interference figure, if the similarity is more than 0.7, then {I1, I2, I3, I4, I5} is taken as interference figure data and input into formula 1; if the similarity is less than 0.7, then {I1, I’2, I’3, I’4, I’5} is taken as interference figure data and input into formula 2;

[0067] Formula 1 is:

[0068] ,

[0069] Wherein, Phi is the wave aberration;

[0070] Formula 2 is:

[0071] ,

[0072] wherein, PhyLoss is a physical loss function, i.e. the interference Figure Two difference in modulus (Loss Function).

[0073] , , , denotes the derivative of the interference image along the lateral direction of the picture, , , , denotes the derivative of the interference image along the longitudinal direction of the picture.

[0074] Embodiment 1

[0075] The following is a case of measuring a DUV lithography objective lens using the present application.

[0076] The light emitted by the 532 nm measurement laser is coupled into a polarization maintaining single-mode optical fiber. The optical fiber is split into two after the beam splitter, and enters two optical fibers. The two optical fibers are respectively referred to as "test optical fiber" and "reference optical fiber". The energy ratio of the "test optical fiber" to the "reference optical fiber" is 9 to 1. In the test optical fiber, there is an energy adjuster. In the reference optical fiber, there is an energy adjuster and a low-precision piezoelectric phase shifter. The phase shift accuracy of the piezoelectric phase shifter is better than 50 nm (≤ 50 nm). The end face of the test optical fiber is flush with the end face of the reference optical fiber and is fixed on the mask plane of the lithography objective lens, and the two are symmetrical about the optical axis of the lithography objective lens ("left and right", "equal spacing"). The center distance between the end faces of the two optical fibers is 1 mm. The aperture stop is composed of a small aperture with a small aperture and a large aperture with a large aperture. The diameter of the small aperture is 500 nm. The large aperture is a square hole with a side length of 10 mm. The aperture stop is fixed on the chip plane of the lithography objective lens. Among them, the small aperture and the end face of the reference optical fiber are in conjugate imaging relationship; the large aperture and the end face of the test optical fiber are in conjugate imaging relationship. The light emitted by the small aperture and the light emitted by the large aperture interfere with each other to form an interference pattern on the image sensor. The computer drives the piezoelectric phase shifter to shift the phase by π / 2, and records the interference pattern again.

[0077] As Figure 3 shown, the artificial intelligence anti-disturbance wave aberration measurement method of the present application comprises: (1) a training stage to obtain a PIUN network (a Physics-informed U-Net network structure based on physical information, the English word is Physics-informed U-Net, abbreviated as PIUN network); (2) an application stage to quickly realize anti-disturbance lithography objective lens wave aberration calculation.

[0078] The training phase method: in a high-vacuum, high-stability point diffraction wave aberration measurement system, a series of interference patterns and their corresponding wave aberrations are measured. The high vacuum refers to a vacuum degree of 1x10 -7 Pa; the high stability refers to a VC-F level shock absorption environment. Taking a measurement as an example, the interference patterns are {I1, I2, I3, I4, I5}. Among them, I1 represents the interference pattern when the phase of the piezoelectric phase shifter is 0, and the phase difference value of the piezoelectric phase shifter corresponding to the interference pattern of I2 and I1 is 2π / 5; and so on, the phase difference value of the piezoelectric phase shifter corresponding to the interference patterns of adjacent serial numbers is 2π / 5. The formula (denoted as formula 1) for calculating the wave aberration Phi is:

[0079] ,

[0080] The training phase steps, as shown in Figure 2 , include:

[0081] Step one: under the conditions of vacuum (1x10 -7 Pa) and high stability (VC-F), 2000 groups of interference patterns are measured, and each group of interference patterns is denoted as {I k1 , I k2 , I k3 , I k4 , I k5}, wherein k is an integer from 1 to 2000. Based on the 2000 groups of interference patterns, 2000 groups of corresponding objective lens wave aberrations are calculated, and each wave aberration is Phi k .

[0082] Step two: based on the deep learning data set {I k1 , I k2 ; Phi k} constructed in step one, the number is 2000 groups, which is the same as step one.

[0083] Step three: based on the data in step two, 1800 groups of data are training data, and 200 groups of data are test data. Based on Figure 3 , 1800 groups of data are input into the PIUN network for training, and 200 groups of data are used for testing until the PIUN network converges.

[0084] Step four: when the loss function (Loss) of the PIUN network converges to 10 -7 , the PIUN network model parameters obtained at this time are the results of the deep learning training phase. In the application phase, only two interference patterns are input, and the lithography objective lens wave aberration can be obtained.

[0085] The application phase, as shown in Figure 4 . Taking a measurement as an example,Figure 1 Two frames of interference patterns are collected in the system, marked as: the first interference pattern and the second interference pattern. Taking the first interference pattern as the input, the PIUN as the deep network, two sequences of interference patterns {I1, I2, I3, I4, I5} and {I1, I'2, I'3, I'4, I'5} are generated, and I2 in the sequence is taken out to perform a "structural similarity" (SSIM) operation with the second interference pattern. If the similarity is greater than 0.7, then {I1, I2, I3, I4, I5} is taken as the interference pattern data and input into formula (1). If the similarity is less than 0.7, then {I1, I'2, I'3, I'4, I'5} is taken as the interference pattern data and input into formula (2) (formula (2) is shown below).

[0086] As shown in Figure 5 , the PIUN network is composed of an image input layer, an encoder, a decoder and an image output layer. The image input layer has only one, and the image output layer has five. Each image output layer is in a parallel relationship. Each image output layer corresponds to an independent encoder and decoder. As shown in Figure 6 , each encoder is composed of 10 encoding modules. As shown in Figure 7 , each decoder is composed of 10 decoding modules. As shown in Figure 8 , each encoding module is composed of a 3x3 convolution module, a normalization module and a Leaky Relu activation module. Similarly, as shown in Figure 9 , each decoding module is composed of a 3x3 deconvolution module, a normalization module and a Leaky Relu activation module.

[0087] The innovative loss function of the PIUN network is shown in Figure 10 . The loss function (LOSS) includes: the MSE loss function (MSELoss) and the PhyLoss physical loss function (PhyLoss). The MSE loss function is consistent with the general MSE loss function definition in the field of artificial intelligence. The PhyLoss physical loss function of the application is constructed by derivative operation of the interference pattern, as follows (denoted as formula 2):

[0088] ,

[0089] I1, I'2, I'3, I'4, I'5 are interference patterns, , , , represent the derivatives of the interference image along the horizontal direction of the picture, , , , represents a derivative of the interferometric image along the picture longitudinal direction.

[0090] The precision test results of the artificial intelligence anti-disturbance based on the application are shown in Figure 11 Figure 11 is the experimental data comparison of the application and the traditional time series piezoelectric phase shift method. The experimental results show that, under the same laboratory conditions, compared with the time phase shift method (five-step phase shift wavefront aberration measurement method, PSI5), the PIUN method improves the RMS repeatability of the obtained wavefront aberration measurement results by 88.7%.

[0091] The specific embodiments of the application are described above with reference to the accompanying drawings, but these descriptions cannot be understood as limiting the scope of the application, the protection scope of the application is defined by the appended claims, and any modification based on the claims of the application is within the protection scope of the application.​

Claims

1. An artificial intelligence anti-disturbance wave aberration measurement method of a lithography objective, one side of the lithography objective to be measured has a reticle surface, and the opposite side serves as a chip surface, characterized in that, The lithography objective wave aberration measurement device comprises: a laser for emitting measuring light; a beam splitter connected to the laser through an incident optical fiber, and also connected to a test optical fiber and a reference optical fiber, for splitting the light emitted by the laser into two beams, one entering the test optical fiber as test light and the other entering the reference optical fiber as reference light; a piezoelectric phase shifter arranged in the reference optical fiber for phase-shifting the reference light; a pinhole diaphragm having a small hole and a large hole, the small hole being circular and the large hole being a regular polygon or circular; an image sensor for detecting and recording interference patterns; The AI anti-disturbance lithography objective wave aberration measurement method comprises the following steps: (1) The end faces of the test optical fiber and the reference optical fiber are leveled and fixed on the mask plane of the lithography objective to be measured, and the positions fixed on the mask plane are symmetrical about the optical axis of the lithography objective to be measured; the pinhole diaphragm is fixed on the chip plane of the lithography objective to be measured, and the small hole and the end face of the reference optical fiber are in conjugate imaging relationship, the large hole and the test optical fiber are in conjugate imaging relationship, the light emitted from the small hole and the light emitted from the large hole form an interference pattern on the image sensor, which is recorded as a first interference pattern; the piezoelectric phase shifter is controlled to shift by π / 2, which is recorded as a second interference pattern; (2) A PIUN network is trained, which is a U-shaped neural network structure based on physical information; In step (2), a series of interferograms and corresponding wavefront aberrations are measured, the interferograms are denoted as {I k1 , I k2 , I k3 , I k4 , I k5}, and the wavefront aberrations corresponding to the interferograms are denoted as Phi k , where k takes a value from 1 to the total number of the series of interferograms; Constructing deep learning dataset{I k1 , I k2 ; Phi k}, the deep learning dataset is divided into training data and validation data, the training data is input into the initial PIUN network for training, and the validation data is tested until the PIUN network converges, when the loss function of the PIUN network converges to 10 -7 , the trained PIUN network is obtained; (3) The first interference pattern is input into the PIUN network to generate two sequences of interference patterns {I1, I2, I3, I4, I5} and {I1, I'2, I'3, I'4, I'5}, in which I1 represents the interference pattern when the phase of the piezoelectric phase shifter is 0, and the phase difference between the piezoelectric phase shifter corresponding to the subsequent interference pattern and the previous interference pattern is 2π / 5; I2 in the sequence is subjected to structural similarity operation with the second interference pattern, if the similarity exceeds 0.7, {I1, I2, I3, I4, I5} is taken as interference pattern data and input into formula 1; if the similarity is less than 0.7, {I1, I'2, I'3, I'4, I'5} is taken as interference pattern data and input into formula 2; Formula 1 is: ， wherein Phi is the wave aberration; Formula 2 is: ， wherein, PhyLoss is a physical loss function, , , , denotes the derivative of the interference image along the horizontal direction of the picture, , , , denotes the derivative of the interference image along the vertical direction of the picture.

2. The method for measuring wavefront aberration of lithography objectives with artificial intelligence-resistant perturbation as described in claim 1, characterized in that, The measurement is carried out in a high-vacuum and high-stability point diffraction wave aberration measurement system, the high vacuum refers to a vacuum degree ≤1×10 -7 Pa, and the high stability refers to a VC-F level shock absorption environment.

3. The method of claim 1, wherein the method is an artificial intelligence (AI) -resistant wavefront aberration measurement method for a lithography objective lens. The PIUN network is composed of an image input layer, an encoder, a decoder and an image output layer, the image input layer is 1, the image output layer is 5, each image output layer is in parallel relationship, and each image output layer corresponds to an independent encoder and decoder.

4. The method of claim 3, wherein the method further comprises: Each encoder is composed of 10 encoding modules, and each decoder is composed of 10 decoding modules, Each encoding module is composed of a 3×3 convolution module, a normalization module and a Leaky Relu activation module; Each decoding module is composed of a 3×3 deconvolution module, a normalization module and a Leaky Relu activation module.

5. The method for measuring wavefront aberration of lithography objectives with artificial intelligence-resistant perturbation according to claim 1, characterized in that, Each of the test optical fiber and the reference optical fiber is provided with an energy adjuster for adjusting the energy of light.

6. The method of claim 5, wherein the method further comprises: The energy ratio of the reference light to the test light is 1:7-11 after the beam splitter splits the light.

7. The method of claim 1, wherein the method is an artificial intelligence (AI) -resistant wavefront aberration measurement method for a lithography objective lens. The phase shift accuracy of the piezoelectric phase shifter is less than or equal to 50 nm.

8. The method for measuring wavefront aberration of lithography objectives with artificial intelligence-resistant perturbation according to claim 1, characterized in that, The diameter of the small hole is 50%-95% of the wavelength value of the light emitted by the laser; when the large hole is a regular polygon, the inscribed circle area of the regular polygon is greater than or equal to the pupil area of the to-be-tested photolithography objective lens; when the large hole is a circle, the circle area is greater than or equal to the pupil area of the to-be-tested photolithography objective lens.

Images