A holographic measurement method and system based on virtual-real combined closed-loop feedback

Abstract

The present application relates to a kind of holographic measurement method and system based on virtual-real combination closed loop feedback, comprising the following steps: obtaining the initial actual hologram of sample to be measured;Based on actual hologram, respectively using denoising network, aberration identification network obtains the hologram after purification, aberration type label;Using the digital holographic model of pre-constructed sample to be measured and aberration coefficient, corresponding simulation hologram is generated;By angle spectrum reconstruction to the hologram after purification and simulation hologram, actual preliminary phase and simulation phase are obtained and phase difference is calculated, and aberration coefficient is obtained by fitting;Based on aberration coefficient, hologram compensation and / or sample to be measured posture adjustment are carried out, and a round of feedback is completed;Actual hologram and / or simulation hologram are updated, and the next round of feedback is carried out, until phase difference meets preset condition;Three-dimensional topography measurement result is obtained, and digital holographic model is updated based on three-dimensional topography measurement result, and holographic measurement is completed.

Description

Technical Field

[0001] This invention relates to the fields of optical measurement and semiconductor manufacturing inspection technology, and in particular to a holographic measurement method and system based on virtual-real combined closed-loop feedback. Background Technology

[0002] In recent years, optical interferometry and digital holography have been introduced into the field of microscopic topography measurement due to their advantages of non-contact, full-field, and rapid acquisition of phase information. However, when applied to real and complex industrial scenarios, existing technical solutions face fundamental bottlenecks: First, unavoidable mixed noise in the detection environment, such as speckle and Gaussian noise, severely reduces the signal-to-noise ratio of the hologram, affecting reconstruction accuracy; second, inherent aberrations of optical systems, such as defocus and astigmatism, are amplified in nanoscale measurements, introducing system errors that are difficult to eliminate; more importantly, the traditional measurement process is an open-loop system, meaning that once the above error sources are introduced, they cannot be dynamically identified and compensated during the measurement process, ultimately limiting the measurement accuracy to inherent hardware and environmental defects, preventing further breakthroughs.

[0003] Chinese patent application publication number CN109029290A discloses a coaxial optical path three-dimensional microscopic topography measurement system. By using a reference objective lens to adjust the optical path in digital holographic microscopy, it solves the problems of reduced noise resistance and aberration caused by the complexity of the optical path, and realizes a high-precision, portable coaxial optical path three-dimensional microscopic topography measurement system, improving the system's noise resistance and measurement accuracy. However, it is still an open-loop system in reality, and the measurement accuracy is limited.

[0004] In summary, there is currently a lack of a holographic measurement method and system to solve or partially solve the aforementioned problems. Summary of the Invention

[0005] The purpose of this invention is to overcome the defects of the prior art by providing a holographic measurement method and system based on virtual-real combined closed-loop feedback, so as to solve or partially solve the problem of unsatisfactory measurement accuracy of existing open-loop schemes.

[0006] The objective of this invention can be achieved through the following technical solutions: One aspect of the present invention provides a holographic measurement method based on a virtual-real combined closed-loop feedback, characterized by comprising the following steps: Obtain the initial actual hologram of the sample to be tested; Based on the actual hologram, a denoising network and an aberration recognition network are used to obtain the purified hologram and aberration type labels, respectively. Using a pre-constructed digital holographic model of the sample to be tested, a corresponding simulated hologram is generated; The pre-acquired initial compensation matrix is ​​superimposed on the phase plane corresponding to the purified hologram to complete the system error pre-compensation. The angular spectrum is reconstructed on the pre-compensated hologram and the simulated hologram to obtain the actual preliminary phase and the simulated phase and calculate the phase difference. Based on the phase difference, the aberration coefficients are obtained by fitting with Zernike polynomials. During the fitting process, the coefficient terms corresponding to the aberration type label are activated. Holographic compensation and / or sample pose adjustment are performed based on the aberration coefficients and aberration type labels to complete one round of feedback. Update the actual hologram and / or simulated hologram, and perform the next round of feedback until the magnitude of the aberration coefficient is less than the preset threshold; The three-dimensional topography measurement results are obtained, and the digital holographic model is updated based on the three-dimensional topography measurement results to complete the holographic measurement.

[0007] As a preferred technical solution, the process of generating the corresponding simulated hologram includes: Based on the laser wavelength, objective lens numerical aperture, theoretical distance between the sample and the camera, and preset aberration coefficients, the corresponding simulated hologram and simulated morphology are obtained using a digital holographic model of the sample under test based on angular spectrum propagation.

[0008] As a preferred technical solution, the process of obtaining aberration coefficients based on phase difference fitting includes the following steps: Based on the phase difference, a set of coefficient vectors is obtained by fitting Zernike polynomials, where low-order coefficients represent tilt and defocus, and high-order coefficients correspond to the aberration type labels, including astigmatism, coma, spherical aberration, and higher-order non-rotational symmetric aberrations.

[0009] As a preferred technical solution, the hologram compensation includes the following steps: The aberration coefficients are converted into a phase compensation surface, and angular spectrum reconstruction is performed based on the phase compensation surface.

[0010] As a preferred technical solution, the attitude adjustment of the sample to be tested includes the following steps: If the aberration type label is tilt aberration, the low-order coefficients representing the overall tilt in the aberration coefficients are converted into voltage signals to drive the action structure to fine-tune the sample posture, eliminate the tilt component, and realize posture adjustment. If the aberration type label is defocus / astigmatism / coma / spherical aberration, the aberration coefficient is converted into a phase compensation factor to perform phase compensation in the actual hologram and / or angular spectrum reconstruction process.

[0011] As a preferred technical solution, the aberration type label is used for: As a prior constraint for Zernike polynomial fitting, only the coefficients corresponding to the aberration type are activated during fitting. As a basis for selecting feedback control strategies, tilt aberrations and wavefront aberrations are distinguished and driven respectively for sample attitude adjustment and hologram phase compensation.

[0012] As a preferred technical solution, the process of obtaining the initial compensation matrix includes the following steps: Before measurement, a standard height step is used as a reference sample. The reference sample is measured to obtain a hologram. Simulations were performed using a theoretical model of a standard sample; By comparison, the initial system error and reference aberration are calculated, and the initial compensation matrix is ​​constructed.

[0013] As a preferred technical solution, the denoising network is a U-Net network with an embedded hybrid attention mechanism, which includes channel attention and spatial attention. An attention module is inserted after each downsampling layer of the U-Net encoder to weight and enhance the interference fringe feature channels and the effective spatial region, thereby suppressing noise and stray interference. The aberration recognition network is a ResNet network.

[0014] As a preferred technical solution, the process of acquiring the initial actual hologram includes the following steps: The laser beam emitted from the laser source is split into an object beam and a reference beam by the first beam splitter; The objective light is irradiated onto the surface of the sample to be tested, which is placed on a precision piezoelectric sample stage, through a microscope objective lens, and carries the surface morphology information of the sample after reflection. The reference light is reflected by the reference mirror; The light reflected from the surface of the sample under test and the light reflected from the reference mirror converge at the second beam splitter and interfere. The interference pattern is captured by a CMOS camera to obtain the initial actual hologram.

[0015] Another aspect of the present invention provides a holographic measurement system based on virtual-real combined closed-loop feedback for implementing the aforementioned holographic measurement method, the system comprising: The physical measurement unit, including the Mach-Zehnder interferometer optical path, is used to acquire the initial actual hologram of the sample under test; The virtual simulation unit has a built-in digital holographic model based on angular spectrum propagation, which is used to obtain the simulated hologram and simulated morphology of the sample under test. The feedback control unit is used to obtain a purified hologram and aberration type labels based on the actual hologram using a U-Net denoising network with embedded hybrid attention mechanism and a ResNet aberration recognition network, respectively. It then superimposes an initial compensation matrix onto the phase plane corresponding to the purified hologram to complete system error pre-compensation. By reconstructing the angular spectrum of the purified hologram and the simulated hologram, it obtains the actual preliminary phase and the simulated phase and calculates the phase difference. Based on the aberration type labels, it constrains Zernike polynomial fitting to obtain aberration coefficients. It selects a feedback strategy according to the aberration type labels and performs hologram compensation and / or sample pose adjustment based on the aberration coefficients to complete the feedback.

[0016] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) By constructing a digital holographic model that is synchronous and parallel to the physical measurement system, a digital twin measurement architecture combining virtual and real elements is formed; a deep learning network with an embedded hybrid attention mechanism is used to achieve efficient denoising and intelligent aberration recognition of holograms; using aberration type labels as prior constraints and feedback strategies, combined with Zernike polynomial fitting and closed-loop iterative correction, phase compensation at the software level and sample attitude fine-tuning at the hardware level are achieved simultaneously; and an initial compensation matrix is ​​obtained through standard sample calibration to further suppress system errors. This invention can dynamically compensate for optical aberrations, attitude errors, and system noise, significantly improving the accuracy, robustness, and adaptability of digital holographic three-dimensional topography measurement, and can achieve high-precision, full-field, fast, and non-destructive measurement of micro-nano structures.

[0017] (2) By constructing a digital holographic model (virtual digital twin system) that is synchronous and parallel with the physical measurement system, artificial intelligence is used to perform signal purification and aberration recognition on the actual holograms collected. On this basis, a comparison and closed-loop feedback mechanism between the virtual and physical systems is established. This mechanism can dynamically compensate for system errors and optimize measurement parameters, thereby breaking through the inherent accuracy limit of the physical measurement system and finally completing the sub-nanometer precision, full-field, fast and non-destructive measurement of the three-dimensional morphology. Attached Figure Description

[0018] Figure 1 This is a flowchart of the holographic measurement method based on virtual-real combined closed-loop feedback in the embodiment; Figure 2 This is a schematic diagram of the Mach-Zehnder interferometric digital holographic optical path principle in the embodiment; Figure 3 This is a schematic diagram of the U-Net denoising network incorporating an attention mechanism in the embodiment. Figure 4 This is a schematic diagram of the ResNet-based aberration classification network in the embodiment. Among them, 1. Monochromatic laser; 2. Lens group focusing; 3. Aperture; 4. Lens group collimation; 5. Lens group beam expander; 6. Test object; 7. Microscope objective; 8. Beam splitter; 9. Charge-coupled device (CCD); 10. Imaging lens; 11. Mirror; 12. Piezoelectric ceramic actuator; 13. Control platform; 14. Computer processing unit. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] Example 1 To address the aforementioned problems in the existing technology, this embodiment provides a holographic measurement method based on virtual-real combined closed-loop feedback. It aims to overcome the three core bottlenecks in existing semiconductor nanoscale topography measurement technology: slow measurement speed, weak anti-interference ability, and limited accuracy of open-loop systems. This method enables a self-calibrating, self-optimizing, and highly robust intelligent measurement system, providing a new precision measurement tool for semiconductor process quality control and yield improvement.

[0021] See Figure 1 The method includes the following steps: Step S1: System initialization, calibration of the physical measurement unit, and synchronization of calibration parameters to the virtual simulation unit.

[0022] The system is started, and the physical measurement unit is calibrated using a standard height sample. The calibration parameters are then synchronized to the virtual simulation unit to build a virtual-real environment with consistent initial conditions.

[0023] Step S2: Acquire the initial actual hologram of the sample to be tested through the physical measurement unit.

[0024] The physical measurement unit measures the sample to obtain an initial actual hologram containing noise and system aberrations.

[0025] Specifically, the sample is placed on the sample stage, and the physical measurement unit takes a picture to obtain the initial actual hologram H_actual_0.

[0026] Step S3: The initial actual hologram is input into the intelligent feedback control unit for processing, and the corresponding simulated hologram is generated by the virtual simulation unit.

[0027] Specifically, on one hand, H_actual_0 is input into the processing module, where the denoising sub-network processes and outputs the purified hologram H_actual_denoised_0. The aberration recognition sub-network analyzes H_actual_denoised_0 and outputs a preliminary aberration type label.

[0028] See Figure 3 This diagram illustrates a denoising network. It first performs multi-level encoding downsampling on the input features, with each level including a 3x3 convolution, ReLU activation, and pooling. This is followed by the same number of levels of decoding upsampling, with each level including a transposed convolution, a 3x3 convolution, and ReLU activation. Skip connections are used between each upsampling and downsampling level, and the result of the final decoding upsampling level is the output feature map.

[0029] A hybrid attention mechanism combining channel attention and spatial attention is employed, with a hybrid attention module inserted at the end of each downsampling block in the UNet encoder. The channel attention branch weights the feature map's channel dimensions, enhancing the relevant channels of the holographic interference fringes. The spatial attention branch weights the feature map's spatial dimensions, suppressing speckle noise and background clutter. The attention-weighted features are then fed into the decoder for upsampling and skip connections. By inserting an attention module after each downsampling layer in the UNet encoder, the interference fringe feature channels and the effective spatial region are weighted and enhanced, suppressing noise and spurious interference.

[0030] On the other hand, the virtual simulation unit, based on the current system parameters including the initial compensation C0, assumes the sample surface to be an ideal plane and generates a simulation hologram H_sim_0.

[0031] Step S4: Through the error analysis module, the actual hologram reconstruction result after processing is compared with the simulation result corresponding to the simulated hologram, and the system error vector is quantified.

[0032] Specifically, the virtual-real comparison module reconstructs H_actual_denoised_0 and H_sim_0 using the angular spectrum method to obtain the actual preliminary phase φ_actual_0 and the simulated phase φ_sim_0. The phase difference between the two is calculated as Δφ0 = φ_actual_0 - φ_sim_0.

[0033] Fitting Δφ0 with the first 15 Zernike polynomials yields a coefficient vector E0 = [a1, a2,..., a...]. 15 ], where a1, a2, a3 usually represent tilt and out of focus, a4-a 15 This represents higher-order aberrations such as astigmatism and coma.

[0034] The feedback optimization decision module determines whether the magnitude of E0 is greater than the threshold ε. Since this is the first measurement, E0 will inevitably be large, triggering feedback.

[0035] Step S5: The feedback optimization module determines whether the system error vector exceeds the threshold. If so, closed-loop feedback is executed. Closed-loop feedback includes: feeding the system error vector back to the virtual simulation unit to update its model; and generating feedback instructions to compensate for the measurement data of the physical measurement unit or adjust its hardware parameters.

[0036] Specifically, the feedback includes: a. Feedback to the virtual simulation unit: Update the internal model parameters of the virtual simulation unit using error data to make it more accurately match the current real state of the physical system.

[0037] b. Feedback to the physical measurement unit: Perform software compensation: Convert error vectors, such as Zernike aberration coefficients, into phase compensation factors, which are then directly applied to correct the initial actual hologram or reconstruction algorithm.

[0038] Hardware adjustment: The mechanically correctable portion of the error vector is converted into a control signal to drive the sample stage or micromirror for fine-tuning.

[0039] Specifically, the virtual-real comparison module reconstructs H_actual_denoised_0 and H_sim_0 using the angular spectrum method to obtain the actual preliminary phase φ_actual_0 and the simulated phase φ_sim_0. The phase difference between the two is calculated as Δφ0 = φ_actual_0 - φ_sim_0.

[0040] Fitting Δφ0 with the first 15 Zernike polynomials yields a set of coefficient vectors E0 = [a1, a2,..., a15], where a1, a2, and a3 typically represent tilt and defocus, while a4-a15 represent higher-order aberrations such as astigmatism and coma.

[0041] The feedback optimization decision module determines whether the magnitude of E0 is greater than the threshold ε. Since this is the first measurement, E0 will inevitably be large, triggering feedback.

[0042] Step S6: Based on the system parameters after feedback, perform the next measurement or data iteration processing, and repeat steps S3 to S5 until the system error vector meets the convergence condition.

[0043] Specifically, the data is either compensated by software or re-collected after hardware fine-tuning. In this embodiment, software compensation iteration is preferred as a new starting point.

[0044] In the nth iteration, the processing module denoises the compensated hologram. The virtual simulation unit generates H_sim_n using the updated model. A comparison is performed again to calculate a new, smaller error vector E_n. This loop continues until, in the Nth iteration, ||E_N|| < ε, at which point the system is considered to have converged.

[0045] Step S7: Output the final three-dimensional topography measurement results.

[0046] When the system reaches a stable convergence state, it outputs the final high-fidelity corrected three-dimensional morphology measurement results and simultaneously updates the digital twin model of the sample in the virtual simulation unit.

[0047] Specifically, once the system converges, it outputs a final high-precision 3D topography P_final(x, y), which has minimized the influence of system errors and environmental noise. Simultaneously, P_final(x, y) is used as the true topography of the sample interface and updated in the virtual simulation unit's database to generate a high-fidelity digital twin model of the sample, which can be used for subsequent process simulation or reliability analysis.

[0048] Preferably, after measuring every 10 samples, the sample stage is driven to move the standard stepped sample into the field of view for a rapid measurement. The measurement result is compared with the ideal virtual representation of the standard sample stored in the virtual simulation unit, generating a small drift error vector ΔE_drift. This vector is not used to compensate for the current measurement, but rather to slowly update the underlying aberration model of the virtual simulation unit and the input preprocessing parameters of the processing module, thereby adapting to the slow changes in the laboratory environment.

[0049] In summary, this method includes training data generation and dynamic calibration mechanisms. The virtual simulation unit is crucial for generating training data. By changing the simulated sample morphology (e.g., random rough surfaces, steps, grooves), adding different levels of mixed noise models such as weighted combinations of Poisson noise, Gaussian noise, and speckle noise, and introducing different combinations of Zernike aberrations, a large number of simulated hologram-ideal phase pairs are generated in batches for training the denoising network and aberration recognition network.

[0050] Example 2 Based on Example 1, this example provides a holographic measurement system based on virtual-real combined closed-loop feedback to implement the holographic measurement method of Example 1. The system includes: (1) Entity measurement unit.

[0051] It includes a Mach-Zehnder interferometer optical path, used to obtain the initial actual hologram of the sample under test.

[0052] For details, see Figure 2In this embodiment, the physical measurement unit employs a Mach-Zehnder interferometry optical path. A laser beam (wavelength λ = 632.8 nm) is split into an object beam and a reference beam by a beam splitter. The object beam is irradiated onto the sample surface placed on a precision piezoelectric sample stage through a 20x microscope objective (NA = 0.4), and after reflection, carries information about the sample surface morphology. The reference beam is reflected by a reference mirror. The two beams converge and interfere at a second beam splitter. The interference pattern is captured by a scientific-grade CMOS camera (pixel size 6.5 μm × 6.5 μm, resolution 2048 × 2048) and transmitted to a computer to form an actual hologram H_actual(x, y).

[0053] (2) Virtual simulation unit.

[0054] It has a built-in digital holographic model based on angular spectrum propagation, which is used to obtain the simulated hologram and simulated morphology of the sample under test.

[0055] Specifically, the virtual simulation unit is a software module running on the same computer. This unit incorporates a digital holographic model based on angular spectrum propagation theory. The model's input parameters include: laser wavelength (λ), objective lens numerical aperture (NA), theoretical distance between the sample and the camera (d0), and preset aberration coefficients (such as Zernike polynomial coefficients). Given a hypothetical sample digital model M_virtual(x, y), which can initially be planar, this unit can simulate and generate the corresponding simulated hologram H_sim(x, y) and simulated topography P_sim(x, y).

[0056] (3) Feedback control unit.

[0057] This method is used to obtain purified holograms and aberration type labels based on actual holograms using denoising networks and aberration recognition networks, respectively. By reconstructing the angular spectrum of the purified holograms and simulated holograms, the actual preliminary phase and simulated phase are obtained and the phase difference is calculated. Aberration coefficients are obtained by fitting, and hologram compensation and / or attitude adjustment of the sample under test are performed based on the aberration coefficients to complete the feedback.

[0058] Specifically, the feedback control unit is also a software module. The processing module includes a pre-trained U-Net denoising network incorporating an attention mechanism and a ResNet-based aberration classification network. The virtual-real comparison and error analysis module is responsible for calculating the difference between H_actual and H_sim, or their reconstructed phases. The feedback optimization decision module generates compensation parameters or control commands based on the difference.

[0059] Before measurement, system initialization is required, using a standard height step (200 nm) as a reference sample. The physical measurement unit measures this standard sample to acquire a hologram. The virtual simulation unit simulates the standard sample using its theoretical model. The feedback control unit calculates the initial system error and reference aberration, such as overall tilt, through comparison, and applies this initial compensation matrix C0 to all subsequent data processing flows to ensure that the virtual and physical systems start working from the same baseline.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A holographic measurement method based on virtual-real combined closed-loop feedback, characterized in that, Includes the following steps: Obtain the initial actual hologram of the sample to be tested; Based on the actual hologram, a denoising network and an aberration recognition network are used to obtain the purified hologram and aberration type labels, respectively. Using a pre-constructed digital holographic model of the sample to be tested, a corresponding simulated hologram is generated; The pre-acquired initial compensation matrix is ​​superimposed on the phase plane corresponding to the purified hologram to complete the system error pre-compensation. The angular spectrum is reconstructed on the pre-compensated hologram and the simulated hologram to obtain the actual preliminary phase and the simulated phase and calculate the phase difference. Based on the phase difference, the aberration coefficients are obtained by fitting with Zernike polynomials. During the fitting process, the coefficient terms corresponding to the aberration type label are activated. Holographic compensation and / or sample pose adjustment are performed based on the aberration coefficients and aberration type labels to complete one round of feedback. Update the actual hologram and / or simulated hologram, and perform the next round of feedback until the magnitude of the aberration coefficient is less than the preset threshold; The three-dimensional topography measurement results are obtained, and the digital holographic model is updated based on the three-dimensional topography measurement results to complete the holographic measurement.

2. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The process of generating the corresponding simulated hologram includes: Based on the laser wavelength, objective lens numerical aperture, theoretical distance between the sample and the camera, and preset aberration coefficients, the corresponding simulated hologram and simulated morphology are obtained using a digital holographic model of the sample under test based on angular spectrum propagation.

3. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The process of obtaining aberration coefficients based on phase difference fitting includes the following steps: Based on the phase difference, a set of coefficient vectors is obtained by fitting Zernike polynomials, where low-order coefficients represent tilt and defocus, and high-order coefficients correspond to the aberration type labels, including astigmatism, coma, spherical aberration, and higher-order non-rotational symmetric aberrations.

4. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The hologram compensation includes the following steps: The aberration coefficients are converted into a phase compensation surface, and angular spectrum reconstruction is performed based on the phase compensation surface.

5. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The attitude adjustment of the sample to be tested includes the following steps: If the aberration type label is tilt aberration, the low-order coefficients representing the overall tilt in the aberration coefficients are converted into voltage signals to drive the action structure to fine-tune the sample posture, eliminate the tilt component, and realize posture adjustment. If the aberration type label is defocus / astigmatism / coma / spherical aberration, the aberration coefficient is converted into a phase compensation factor to perform phase compensation in the actual hologram and / or angular spectrum reconstruction process.

6. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The aberration type label is used for: As a prior constraint for Zernike polynomial fitting, only the coefficients corresponding to the aberration type are activated during fitting. As a basis for selecting feedback control strategies, tilt aberrations and wavefront aberrations are distinguished and driven respectively for sample attitude adjustment and hologram phase compensation.

7. The holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The process of obtaining the initial compensation matrix includes the following steps: Before measurement, a standard height step is used as a reference sample. The reference sample is measured to obtain a hologram. Simulations were performed using a theoretical model of a standard sample; By comparison, the initial system error and reference aberration are calculated, and the initial compensation matrix is ​​constructed.

8. A holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The denoising network is a U-Net network with an embedded hybrid attention mechanism, which includes channel attention and spatial attention. An attention module is inserted after each downsampling layer of the U-Net encoder to weight and enhance the interference fringe feature channels and the effective spatial region, thereby suppressing noise and spurious interference. The aberration recognition network is a ResNet network.

9. A holographic measurement method based on virtual-real combined closed-loop feedback according to claim 1, characterized in that, The process of obtaining the initial actual hologram includes the following steps: The laser beam emitted from the laser source is split into an object beam and a reference beam by the first beam splitter; The objective light is irradiated onto the surface of the sample to be tested, which is placed on a precision piezoelectric sample stage, through a microscope objective lens, and carries the surface morphology information of the sample after reflection. The reference light is reflected by the reference mirror; The light reflected from the surface of the sample under test and the light reflected from the reference mirror converge at the second beam splitter and interfere. The interference pattern is captured by a CMOS camera to obtain the initial actual hologram.

10. A holographic measurement system based on virtual-real combined closed-loop feedback, characterized in that, For implementing the holographic measurement method as described in any one of claims 1-9, the system comprises: The physical measurement unit, including the Mach-Zehnder interferometer optical path, is used to acquire the initial actual hologram of the sample under test; The virtual simulation unit has a built-in digital holographic model based on angular spectrum propagation, which is used to obtain the simulated hologram and simulated morphology of the sample under test. The feedback control unit is used to obtain a purified hologram and aberration type labels based on the actual hologram using a U-Net denoising network with embedded hybrid attention mechanism and a ResNet aberration recognition network, respectively. It then superimposes an initial compensation matrix onto the phase plane corresponding to the purified hologram to complete system error pre-compensation. By reconstructing the angular spectrum of the purified hologram and the simulated hologram, it obtains the actual preliminary phase and the simulated phase and calculates the phase difference. Based on the aberration type labels, it constrains Zernike polynomial fitting to obtain aberration coefficients. It selects a feedback strategy according to the aberration type labels and performs hologram compensation and / or sample pose adjustment based on the aberration coefficients to complete the feedback.

Images